[00:00:05.49] PRESENTER: Take it away. 

[00:00:06.30] YUE CHEN: Thank you. Thank you so much for the great introduction. So today I will just briefly explain what we have been done in the past several years. The topic of the presentation is Tentacle-like Continuum Robots for Minimally-Invasive Surgery. Actually, so when we talk about robot, our first impression might be this, this giant industrial robot arm, and put it in the assembly line to perform different tasks. 

[00:00:33.15] These industrial robots are very, very precise, and they can be used to handle different heavy load tasks-- for example, welding and lifting some objects. But as you can imagine, this industry robot is very dangerous. Right? Because they have been operated inside the protective cage, the working along this industrial robot can be very stressful and even dangerous. In some scenarios, we hope that the robot and human beings can have some safe interaction, just like this. 

[00:01:03.29] Right? So Baymax is holding Hiro. So a [INAUDIBLE] in robotics there are two fundamental approaches to enable safe human-robot interaction. The first approach is called interactive safety. Basically, the different control algorithm or contact force detection algorithm or [INAUDIBLE] algorithm can be developed such that the robot can behave safely. For example, when the robot is in contact with a human, we can detect this condition from that. So these are called interactive safety. 

[00:01:35.73] The second approach is intrinsic safety. So this is a support bot developed by a Japanese company. Different from the conventional robot, this robot is [? fabricated ?] with some softer materials such that they can be safely interact with other people or keep them stay mentally engaged. So continuum robot can be considered as [? technical ?] from the materials perspective. 

[00:01:59.45] So different [INAUDIBLE] have been developed in the past probably 10 to 20 years, and here you can see a continuum robot that can be used for search and rescue. And this is a continuum robot that can be used for grasping and also even used to twist the screw, and this is a continuum [INAUDIBLE] robot and this is a continuum robot for the endoscopic procedures. Different from the conventional robot that is controlled by the joint, the motion of the continuum robot are generated by [INAUDIBLE] deform the robot itself because the robots are fabricated with some [INAUDIBLE] form of materials like silicone or nanotubes, nanomaterials. 

[00:02:41.90] What I'm particularly interested in using continuum is in the medical field. The main motivation is the ongoing trend to make the procedure less invasive. Here are actually some images from the internet shows the scar of the procedure, and I want to show you how robots can change the whole story by using the novel techniques. 

[00:03:03.64] Here is the open-brain surgery, and, as you can see, this scar is around 10 to 15 inches. But now we can perform the same procedure [INAUDIBLE] such that we can enter into the brain from a minimally-invasive approach. And here this is another probably 15 to 20 inches scar after the open abdominal procedure, but now we can use the da Vinci robot to enter into the abdomen to perform some procedures from the natural opening. 

[00:03:31.48] Nowadays, we not only care about whether the patient can live but also care how the patients feel about themselves. So I'm particularly interested in using continuum robot in the medical field. In our lab, we developed different types of continuum robots, and here is a concentric robot [INAUDIBLE] just mentioned, and the second one is catheter. And here is the catheter maneuver into the left atrium for the ablation, and the third one is soft robot. And we also developed some non-continuum robot like MRI-guided brachytherapy, and injection needle, and robot-assisted HIFU. 

[00:04:09.35] Today, I will just focus on how to use continuum robot to save the patients' lives. The first continuum robot I want to introduce is MRI-compatible concentric tube robot. Here are my clinical collaborators. So, what is concentric tube robot? The basic concentric tube robot is a group of tubes that are nested together. 

[00:04:30.25] These tubes are pre-shaped at different configurations, and each tube can be linearly translated and also x rotated. Due to this tube interaction, we incurred a very complex workspace compared to a straight tube. So let's assume if we have a linear tube, the only workspace is just a straight line. But because of this tube is pre-shaped is super elastic, we uncovered a much larger workspace. 

[00:04:54.94] If we can add some gripper at the [INAUDIBLE] we can use to grasp on tissue, or if we can integrate a laser fiber, then we can use it to perform the laser ablation. So what I'm doing in concentric tube robot is to use it for the intracerebral hemorrhage. The intracerebral hemorrhage is a major health care challenge. It occurs when the blood leaks from the ruptured vessel and accumulates inside brain, and this blood will compress normal healthy tissue to cause some complications. 

[00:05:25.51] It is a very common disease. Around 1 in 50 people will have ICH in their lifetime. So I think here we have more than 50 people. Someone will have ICH in their lifetime. Not you? OK. 

[00:05:41.26] So around 40% mortality rate in the first 30 days. It's a very, very deadly disease. So therefore the conventional approach for the ICH is very invasive. So basically the surgeon [INAUDIBLE]. 

[00:05:59.35] OK, so it's very invasive. This will [INAUDIBLE] to remove part of the skull and then take the blood tissue out. As you can see, because of the invasiveness, not all the patients are qualified for this open-brain surgery. So neuroendoscope is an ongoing trend to treat the intracerebral hemorrhage, but there are two limitations. 

[00:06:23.06] First of all, the conventional endoscope is very big. As you can see, this BrainPath has a diameter of around 30 millimeters. Just imagine we insert the endoscope that has a similar dimension of finger. Very invasive, and the second approach lacks the dexterity. Because the conventional endoscope is a straight device, if we want to use this endoscope to reach the boundary of this blood clot, what we do is we need to tilt this endoscope around this entry point. 

[00:06:52.30] Because it's tilting motion, this will cause a cone of destruction because this endoscope will compress the surrounding tissue. If we can detect a blood clot from a very minimally-invasive approach, we can probably save this patient. Here is our approach. 

[00:07:08.13] Let's assume that we have an outer tube. This outer tube is straight tube, and it has a diameter of around 1.5 millimeters. Then we have inner tube, and this inner tube is super-elastic one. It's fabricated with nanomaterials. If we retract this inner tube, it will become straight one, and then we can insert into the brain. But once we deploy, because this inner tube is super-elastic, it can return to its initial shape. So this inner tube can cover the entire blood clot. 

[00:07:39.18] So if we can [? profit ?] designs the inner tube curvature, arc length, and also the incision orientation, we can make sure that the workspace of the inner tube can cover the entire blood clot. This whole idea is very simple. Based on this idea, we built a robot, and as you can see, we use this red gelatin to represent the blood clot, and we use this transparent gelatin to represent the normal healthy tissue. 

[00:08:05.53] So we can see that we take all the blood clot out, but from this video we found that we need to have the image feedback. Right? Because in the bench top test we cut the brain model in half, but in the real experiments we want to see how much we have already evacuated where is the inner tube with respect to the boundary of the blood clot. 

[00:08:28.73] So we considered two different types of imaging modalities. The first one is CT and second one is MRI. Using MRI can bring some advantages. For example, we can see this tiny blood clot inside the brain. It provides high-resolution imaging feedback. 

[00:08:43.03] Secondly, there is no radiation to both patients and also the clinicians. So to design a robot that can be operated inside MRI is quite challenging because the MRI has strong magnetic field, so any device fabricated with the ferromagnetic materials. Like the steel can be very dangerous, and also all the conventional [? DC ?] model whether [INAUDIBLE] [? DC ?] model we cannot use that inside scanner because the imaging quality and also safety are constrained. 

[00:09:13.65] As we know for the robot, typically when you have the robot hardware, which is actually [? in part ?] when you have the sensing path. And also we need to have the control mode [INAUDIBLE]. So next I want to discuss all of the several components one by one. 

[00:09:26.20] First of all is robot hardware. This is our MRI-safe pneumatic motor. The cool part of this motor is that it's fabricated with plastic material, which means there's no magnetic force between the motor and also the MRI scanner. This motor is powered by pneumatic power supply, which means from the working principle point of view it's electromagnetically decoupled from the scanner. 

[00:09:50.73] So here you can find some specifications of this motor. Once we have this motor, we develop some control methods to control the motor position. Here I will just show you the setpoint tracking. 

[00:10:01.05] As you can see, we can achieve sub-millimeter control accuracy for this pneumatic motor. Of course, we also test some bandwidth, and also sine wave tracking, and also railroad tracking. Both can achieve the desired performance. 

[00:10:14.26] And once we have this robot hardware-- once we have the motor, we can build the robot hardware. So here is the very first project I developed during my PhD, but now my PhD student, Anthony, has developed a second prototype we just tested last weekend. This robot has [? really good ?] [INAUDIBLE] because the inner tube has insertion and also rotation. Outer tube has insertion, which means we have three active degrees of freedom, and this robot is mounted on a 2D degree aiming device because this aiming device will provide the pulse of this entry point. So in total it's five degrees freedom. 

[00:10:50.24] And once we have hardware, want to explore the control. So the control of this robot can be achieved with the basic rules of the control. I think everybody has already learned that. To start with, this resolved-rate control. First, we need to identify the desirable trajectory, and then we apply the inverse of the [INAUDIBLE] that can map from the [INAUDIBLE] to the joint space project and then we integrate. We can get the joint space position, and then the most important part is when you have the kinetic and static model such that we can match the joint space position to the corresponding robot and any further space position. 

[00:11:34.06] To develop this static or kinematic model, we rely on the Cosserat rod theory. So it's lengthy derivation, but here I will just show you that some very basic ones. First of all, we need to consider the kinematics, and, second, we also need to consider the mechanics. Lastly, we need to consider the [? constitutional ?] laws. 

[00:11:58.95] If we combine all these fixed differential equations together, we can solve the slender rod subject to the [INAUDIBLE] deformation subject to the external force or the internal force. Here a demo shows that once we apply the force at a tip of the soft rod and also at the body of this continuum arm how the deformation will become. And once we have the hardware, modeling, and also control method, we can validate it in the bench top and also in the MRI scanner. 

[00:12:31.58] So next I want to show you the MRI validation we did before. So this is a hardware, and we're using pneumatic air hose to power the robot, and we have the optical fiber. That's how we're going to track the joint space motion, and test the MRI compatibility and also several different experiments to validate the performance in terms of MRI safety. 

[00:12:57.08] Next I want to just show you two critical experiments. First of all is accuracy test. Accuracy test means we have a target inside the MRI image, which is this purple point, and we want to control the robot to hit the target. So here this black part shows the outer tube image after. Because other tube is metallic, it has certain image alpha. As you can see, when we insert this outer tube, we can see that with real-time image feedback we can always see the relative position between the target and also the other tube. 

[00:13:39.52] This task with feedback can be very critical for us to ensure the safe cluster of control and also enhance our confidence level when the surgeon insert in device [INAUDIBLE] based on the immediate feedback. So there were significant [INAUDIBLE] were fairly confident about this procedure. This accuracy test, and the next we want to show is that efficiency test. 

[00:14:01.15] Efficiency test is we want to evaluate how much blood clot we can evacuate during the procedure. Similarly, we use a real-time evacuation monitoring to monitor the whole evacuation outcome. And also a surgeon in a [INAUDIBLE] surgeon can tell how much we have already evacuated and how much we need to continue to evacuate. 

[00:14:21.28] This is a basic experiment we performed before. Last week, we just performed a more comprehensive experiment was still in the process of the data analysis. Our ongoing work includes a robot optimization. This is the most up-to-date robot design, and another ongoing work is the robot [? crossplane, ?] basically when we insert the needle to avoid some critical structure. For example, the DTI and also the arteries inside the brain. 

[00:14:51.14] So considering the dexterity of this continuum concentric tube robot, we can play with a lot of path-planning algorithms. And also we can investigate the optimal design of the concentric tube robot. But we use a three-tube design or two-tube design, or whether it's just a planar circular tube design, or it's a helical shaped design. So this is the ongoing work we are trying to do right now. 

[00:15:17.03] So this is a concentric tube robot. Next I want to talk about the tendon-driven continuum robot. Here are my collaborators. Tendon-driven deliverable have been used in a lot of medical devices. One of the most common medical devices is catheter. So they have been used to [INAUDIBLE] different catheters. For example, intracardiac catheter is the image I want to show you here, and this one is a guiding catheter. 

[00:15:45.25] The deformation of the catheter is very easy. Basically, we have steel knob at the handle, and once we pull it this will cause the deflection of this tip. I say it's very simple mechanical design, but it's very challenging to control because the catheter has dimensions of around more than 1.4 meters and the material is plastic, and also [INAUDIBLE] [? nanowire. ?] It's very challenging to control that. 

[00:16:13.39] In my research, we will focus on the catheter applications in the atrial fibrillation treatment. Here are some information about the atrial fibrillation, and, again, it's a very common disease. 3 to 4 million people are diagnosed every year in the United States. How to treat this disease is, first of all, the surgeon or the cardiologist needs to make an incision at the femoral vein and insert the catheter from the vein although to the heart. 

[00:16:45.80] Once the catheter is inside the heart, the AF ablation will be generated at the tip of this catheter, and this ablation or the thermal energy will kill the cell inside the left atrium chamber such as we can block the irregular electric signal. So this will [INAUDIBLE] atrial fibrillation to make the heart beat in the normal rate. OK? 

[00:17:09.36] But the conventional treatment for the catheter is suboptimal because it has a recurrence rate of around 30% to 50% after first trial, which means around 50% will take probably a second trial or third trial because the recurrence rate is very high. Several different reasons contribute to this. One of the primary reasons is that there is a lack of feedback, lack of very effective feedback to tell the cardiologist where the ablation target is with respect to the catheter based on the fluoroscopy guidance, and it's also very hard for them to tell whether it's acute ablation or the permanent ablation. 

[00:17:50.22] We hope we can create a permanent ablation such that we can effectively block the irregular electrical signal, but sometimes we can only [INAUDIBLE] because the lack of contact force feedback. So MRI can provide a better imaging resolution of the heart, but still this imaging quality is still not very good. What I'm trying to do is-- can we increase the imaging quality of the heart such that we can achieve better feedback? 

[00:18:19.87] As we know, imaging quality and also the coil distance has this reverse relationship. Here you can see different imaging coils inside scan. We had head coil, test coil, abdominal coil, and also leg coil. The coil displacement [INAUDIBLE] the distance between coil to target and also a normalized signal-to-noise ratio has this reverse relationship. If we can place the coil very close to the target, we can get higher image quality. The question becomes, can we bring the imaging coil into the heart? 

[00:18:51.76] Having this basic motivation, we tried to develop an intracardiac imaging catheter. The design requirements include, first of all, this catheter can be folded for the manipulation and it can be used or it can be deployed to cover a large workspace within the heart because basically inside heart we have pretty large catheter or we have large, large space. Want to deploy that so we can imagine a large field of view. 

[00:19:18.15] So we got the idea from umbrella, right? Once we fold it, we can put it in the bag, but once we open it we cover the whole body. Having this basic working principle, we built the first tendon-driven imaging catheter. This is the first prototype, and the second prototype looks like this. 

[00:19:34.37] Once we have this hardware, we test in animal trials in Hopkins, and so these are the experimental setups. And we test in both ex vivo and also in vivo tissues, and here is the result. As we can see, with this intracardiac imaging catheter we can achieve more than five times [INAUDIBLE] signal-to-noise improvement, and it achieves much faster imaging compared to the regular body coils. 

[00:20:01.94] Once we have this imaging catheter, we can achieve the high-resolution imaging feedback. With this high-resolution imaging feedback, we can develop the intra-optical imaging interface such that we can monitor and also plan the procedure. Here this is the left atrium model, and this red circle indicates the desired circular abrasion zone, and this blue rod is the catheter. 

[00:20:28.67] The catheter shape during the procedure can be obtained with a technique called 3D tracking, so we can achieve faster positioning and also high-bandwidth tracking during the procedure. Once this catheter is placed close to the target, we can click the button and say that we have already ablated this region. With this intra-optical imaging feedback, we can always tell how much we have already evacuated. And once [INAUDIBLE] we need how much we have already ablated and what's the rest we need to continue to ablate. This will tell the cardiologist about the completeness of the whole procedure. 

[00:21:09.63] Again, there will be some mathematical model in this. For example, [INAUDIBLE] when the catheter is inside the heart. So when you constrain this motion, it still penetrates into the tissue cause a perforation. Here I want to just show you with this interface what kind of outcome we can get. 

[00:21:28.07] To evaluate this outcome, we want to perform several tests. First is accuracy. We want to show with this interface what's the accuracy improvement we can achieve, and the next is sufficiency of lesion. Basically, we want to characterize how much we have already ablated versus we haven't ablated and also the average procedure time. 

[00:21:48.45] The last safety. The safety is a test of the catheter that exceeds the tissue model and also [INAUDIBLE] tip deviation and also average time. So we invited 10 different subjects to perform the procedure without interface and with the interface, and we can see that with the interface we achieved significant improvement in all the six performance index compared to the one with the interface. On average, subjects demonstrated around 30% improvement in terms of accuracy, efficiency, and safety. This indicates the significant potential for the clinical use of our imaging catheter plus the navigation interface developed based on this imaging catheter and also the modeling such as the [INAUDIBLE] modeling plus the [INAUDIBLE] modeling. 

[00:22:39.68] Next I want to tell you another tendon-driven device inside the tissue. Catheter is the device that is inside the chamber. The chamber is hollow. But now can we deploy a device that can be deflected inside muscle or inside tissue? Here is our application we tried to use for the brachytherapy and this is a clinical background of the gynecologic cancer. 

[00:23:06.08] The truth is this cancer the surgeon will need to use brachytherapy. The basic idea is we insert this straight stylet into the tissue and then removes that because [INAUDIBLE] have the hollowed catheter or hollow tube, then we deploy the RFC into this tube or deploy the [INAUDIBLE] into the tube. Then this radiation will kill the tumor cell. 

[00:23:29.92] As you can see, the most important part for the brachytherapy is how can accurately deploy this [INAUDIBLE] into the patient. Right? If we can deploy at the desired location, then we can kill the tumor. If not, we'll probably kill the normal healthy tissue. 

[00:23:46.58] So to check the device inside tissue, we used active tracking technique. Basically, we mounted some very tiny tracking coil onto the stylet. It's quite similar to the EM checker, but the EM tracker is when you have the field generator [INAUDIBLE] into the scanner. But we used RF tracking. The RF tracking is that we can couple to the MRI scanner wirelessly or wired. 

[00:24:11.14] With this tracking technique, you can say there is corresponding location inside the MRI imaging, and then we'll get this signal quality. Signal of this coil projected on the three axes, for example, xyz, and we can get a peak around each axis. This peak indicates the location of one of the tracking coils in x direction, y direction, and also z direction. Once we have this tracking technique, we can create the device in real time, and we test it in a turkey. 

[00:24:42.74] Here is the needle inserted inside turkey, and we can see the corresponding location of the stylet in real time. We thought it was great. Then we test in the human trials. We just test that, and we test in several different human trials, though we found several problems. First of all, let's analyze this figure. 

[00:25:04.10] This cross indicates starting point, and this start indicates a target. As you can see here, this is the start and this is target. We have a significant error between the starting point and also target. And secondly, because we missed the target, the surgeon will need to insert, retract, insert, retract multiple times in order to hopefully get some tissue-needle interaction. We hit the target. So that's a reason we can say this [INAUDIBLE] insert going back and forth, back and forth. 

[00:25:35.07] So these are the two fundamental problems we found in the human trials. The primary reason is because it lacks the dexterity. Because consider this straight tube. It can only be inserted, right? We are hoping that the target is on the trajectory of this needle, but because tissue deformation and also needle deflection inside tissue there will always be some uncertainties. So that's a reason we couldn't hit the target in a very efficient and also acceptable manner. 

[00:26:07.64] So to kind of increase the dexterity with tendon-driven deflectable stylet, it's very easy to build this hardware. We just make some cut-out on the tube, and then we use [INAUDIBLE] and wire solder here. And once we put in this nanowire, it can deflect. 

[00:26:30.16] This is a deflection of this stylet. Once we have this hardware, we can build this model similar. We use-- reach the [? firm ?] transformation. We can find the corresponding tendon wire displacement caused to the curvature of this deflected region, and we [INAUDIBLE] test this device in the bench top environment but found that there was a huge error. 

[00:27:00.41] There is because this tendon wire not only contributes to the stylet deformation but also contributed to the tendon wire elongation. This is very important. Nobody considered that before because previously we just wanted to make sure this tendon-driven mechanism-- we are hoping that this device is soft, but in our application we need to penetrate into the muscle and penetrate into the tissue. We are [? hoping ?] [INAUDIBLE] develop a study that is very rigid, which means that this tendon wire elongation. We have to consider that. 

[00:27:35.32] Based on this motivation, my student, Anthony, developed energy-based static model. In this new model here, this is a tendon wire contribute to the [INAUDIBLE] deformation, and also this portion is tendon elongation. With this comprehensive model, we validate experimentally, and we can see that without consideration of the tendon elongation this is a [? training ?] angle, but it was a consideration of the elongation. It can match to the experimental data very accurate. 

[00:28:10.73] Then we test it with ex vivo tissue and with EM tracker. We can find that the overall targeting accuracy is around 2 millimeters, which is significantly better compared to our human trial results, and we test in the MRI experiment [INAUDIBLE] inside 1.5 T scanner. We found that overall targeting accuracy is about 2.3 millimeters, which is pretty acceptable because MRI imaging typically has a resolution of around 1 millimeter. So we achieved 2.3 millimeters. Our ongoing work is to get [? RB ?] approval of this developed study and test in the human trials. 

[00:28:53.12] OK. The last part is the bio-inspired soft robot. These are my clinical collaborators. The soft robot-- I know some of you have been working on this soft robot for several years, but initially it was proposed by Suzumori in [INAUDIBLE], but he didn't receive too much attention until probably 10 or 15 years ago 

[00:29:14.63] Dr. Whitesides in Harvard developed a starfish gripper. They can use it to crush the egg in very safe approach. Since then, soft robot has become a very hot topic. A lot of different soft robot have been developed. Here are just some examples. 

[00:29:33.77] In 2015, Daniel and also Michael published a paper in Nature about the design, fabrication, and control of soft robots. In the paper, one of the most important questions they tried to answer or they hope our robotics can answer is how do we design and control these soft machines and how do we use these soft machines. 

[00:29:57.71] This is a very complicated question. To answer this complicated one, we start with the very basic soft robot. So let's consider this one [INAUDIBLE] soft robot. 

[00:30:12.52] Here this is a pneumatic-driven soft robot. Once we pressure that, it will be bent to one direction, and this is a tendon-driven soft robot. Basic idea is quite similar to the catheter. Once we pull the tendon wire, it will deform. 

[00:30:27.43] To fabricate this type of soft robot is very easy. We basically rely on the silicon casting method. We create the soft robot body first and at the base strain-limiting layer such that we can constrain its motion in one direction and also this radical strain-limiting layer such that we can constrain its motion in the vertical direction. And then we add an outer layer to protect both strain-limiting layer, and then we [? fill both ?] [? in. ?] That's the fabrication of this soft robot. 

[00:31:00.61] We can fabricate this soft robot probably within one or two days. Very easy, but to model and just to control the soft robot is very challenging. The conventional approach for the soft robot control, especially for the kinematics in terms of-- from the configuring to the task system mapping, is divided based on this constant curvature assumption. 

[00:31:23.68] With this assumption, we can find there are two problems. First of all is that there is an expression singularity, and, secondly, in reality [? so ?] some soft robot [INAUDIBLE] basic not constant curvature. Right? Just like the robot you're seeing here. 

[00:31:38.38] We cannot just use a part of a circle to represent this non-constant curvature of soft robot. So what we did before is we developed a calibration-based model [INAUDIBLE] also-- can also-- first of all, we found a mapping from the configuration space to task space. 

[00:31:57.70] So let's assume that the configuration space variable is the tangent angle along any point. Then based on this basic geometry, we can find the position with respect to the arc length can be differentiation. It can be written as a tangent vector. The next is to find the actual space q, which is a joint input to the configuration space variable [INAUDIBLE] tangent vector angle When the tangent angle at any point of this soft robot is determined by the input q and also the location. Right? And then we can write this phi, s, and also gamma s into this polynomial representation. 

[00:32:43.93] And what we can do is we can get a family of this soft robot [INAUDIBLE] subject to different input, and then we can put it into a measured form. With this measured form, our goal is to solve the coefficient measured such that we can match the input to the output. To solve this coefficient measure, we can just use the Kronecker product to solve in a least square approach. So this is a solution. 

[00:33:06.87] Now, once we know the actuator space to the configuration space mapping, plus we know the configuration to the [INAUDIBLE], we multiply them together. We can find mapping from the actuator space to the task space. 

[00:33:21.30] I don't know what happened. There's a weird shift here. OK. So what's [INAUDIBLE] this basic model, and we test in the real hardware and we can find out. So the model can match to the experimental results very accurately. 

[00:33:36.52] [INAUDIBLE] know what happened with this one. OK. Having this forward kinematic model, we can develop the different kinematics model and also the inverse very easily using a [INAUDIBLE]. So I'm not going to cover that in great detail. 

[00:33:51.97] Next, we want to talk about stiffness modulation. So why we need to do stiffness modulation? Because we can find a soft robot, sometimes it's quite similar to the octopus arms. It's very soft. It's very hard to hold a heavy object in our experiment, but we know that. So the rigid robot is quite similar to the horse leg, right? 

[00:34:10.64] They're very rigid and very powerful, but the leg dexterity, or the legs that [? combines. ?] So we got the inspiration from the animals, and we are thinking whether we can create some robot that is sometimes soft and sometimes rigid. We found that the tail is meaty, which means it's very soft but sometimes can be very strong. 

[00:34:31.67] Based on this kind of inspiration, we tried to add a backbone to the soft robot. By modulating the backbone into the soft robot, we can control the stiffness. With this basic idea, we create the stiffening modulation with backbone, and first we make a hollow channel inside the soft robot. This black line indicates the backbone, which is fabricated with nitinol such that it can be deformed. 

[00:34:57.41] By inserting this backbone lens, we can modulate the overall stiffness of the soft robot. This a cross-section area of this soft robot. Here is the nitinol backbone insertion, and we create three channels such that we can put the FBG sensor in the soft robot, but we can monitor its shape. 

[00:35:16.87] With this hardware, we need to consider its modeling, so we use a Cosserat modeling, but here it's slightly different because we have the tendon-driven force. So we need to consider the tendon force here, and this is a force when we want to integrate into the previous Cosserat model I just described in the [INAUDIBLE]. With this tendon force, we can say it's information subject to a given force. 

[00:35:44.27] So we developed the model and fabricates hardware. Next we want test experimentally. So this is our experiment setup. We use a linear guided rail to pull this tendon wire. Here we have four sensors such that we can measure the joint space force. 

[00:35:58.07] At the tip, we put a load such that we can provide a constant force, and here this is a FBG sensor such that we can monitor its shape in real time. Were first validated the model accuracy, and then we validated the stiffness modulation. So here I will just show you the [? inner ?] stiffness improve with this backbone. 

[00:36:20.33] The overall stiffness is increased up to 350% compared to the one without the stiffness modulation. So this pink color line indicates the model shape, and this dotted line indicates the external shape. As you can see, our model can match to the experimental results very accurately. The last I want to show you how to use this soft robot in the clinical environment. 

[00:36:47.63] We test this soft robot for the pancreatic cancer treatment. Here is some clinical background. So pancreatic-- so far it is still one of the most deadly cancers in the world. The five-year survival rate is less than 10%. 

[00:37:06.51] So surgical resection of this pancreatic is very invasive. Literally, the needle cuts part of the stomach, and also pancreatic head, and also the bladder. When you cut a lot of organs because the pancreatic cancer is typically located at this region, a lot of very confined area. So what we are trying to do is to use photodynamic therapy to treat this disease. 

[00:37:28.68] The whole idea is that the patient first takes some photosensitive drugs, and then these drugs will be accumulating in the cancerous region compared to the-- it will accumulate more in the cancerous region compared to the normal tissue. Then we expose light to these photosensitive drugs. Because there will be more photosensitized drugs here-- so this photosensitizer will automatically kill the tumor. Problem becomes how can we bring the light the pancreas in a very dexterous approach. 

[00:38:01.17] So here is our solution. This is a conventional laparoscope, and this laparoscope will put a soft robot tip such that this tip can be deformed. And inside this soft robot tip we have very tiny micro-camera such that we can monitor the shape. We can provide a visual survey, and we have the tendon wire such that we can deform the soft robot, and we have photo-- PDT fiber such that we can emit the light from the tip such that we can kill the cancer in this envisioned setup. 

[00:38:36.98] Once we have this basic idea, we developed the hardware, and we experimentally validated. First of all, we validate its accuracy in terms of position accuracy and also the orientation control accuracy. We can find that both accuracy performance can achieve sub-millimeter performance. Then we test in the animal trials, and this blue dot indicates the navigation camera comes from the [? micro-- ?] the navigation camera. 

[00:39:06.26] And this red dots indicates photodynamic therapy laser fiber. As we can see, we can monitor the whole procedure, and we test in animal trials. The performance index involved the tumor dementia. So we measured tumor before the treatment and also 21 days after treatment. We can see the tumor dimensions have been significantly improvement. 

[00:39:31.67] In summary, in this talk I talked about three different types of continuum robots. The first is concentric tube robot, second is tendon-driven robot, last is soft robot. Next I want to acknowledge my collaborators, advisors, and also my students. They did a great job. 

[00:39:49.58] Finally, I want to make an announcement because I'm new here. So we still need to hire a lot of talented students to work on this cool project. So we are looking for students in robotics, mechatronics, control, and design. If you are interested, just let me know. Shoot me an email. I can reply immediately. Thank you. 

[00:40:12.07] [APPLAUSE] 

[00:40:18.15] PRESENTER: We have time for questions [INAUDIBLE]. 

[00:40:23.31] AUDIENCE: [INAUDIBLE] MRI like [INAUDIBLE]. 

[00:40:34.05] YUE CHEN: Uh-huh. 

[00:40:35.27] AUDIENCE: [INAUDIBLE] 

[00:40:38.90] YUE CHEN: So that's a good question. Because there are several problems in there to evaluate safety, first of all, it's whether this device can generate heat. So far, we didn't see any heating problem, and second is whether this device is biocompatible. So we have some protection [INAUDIBLE] cover to protect it. 

[00:41:01.94] So far, we believe it's safe such that it can be deployed into the heart. Good question. Not yet. Basically, for this MRI-guided trial there are not too many groups can perform this procedure in-- 

[00:41:36.05] AUDIENCE: [INAUDIBLE] 

[00:41:40.36] YUE CHEN: I still need to explore, but so far I know Hopkins, Harvard, and Stanford. Those three can do that, so that's the reason we've built a strong connection with Hopkins. But [INAUDIBLE] I'm not quite sure about interventional MRI capability. Probably they can. I just need some time to figure out who can do that. Yeah. 

[00:42:04.31] AUDIENCE: [INAUDIBLE] 

[00:42:05.80] AUDIENCE: So you're able to model the stiffness [INAUDIBLE]. Otherwise, you have to [INAUDIBLE] robots, but I know that soft robots that are [INAUDIBLE]. But were you able to get [INAUDIBLE] modulation [INAUDIBLE]. 

[00:42:38.58] YUE CHEN: So you are asking whether we rely on the visual feedback to get the shape of the soft robot or-- yes. 

[00:42:47.58] AUDIENCE: [INAUDIBLE] 

[00:42:53.90] YUE CHEN: You are asking whether this [INAUDIBLE] we can get to the desired position. Right? So, yeah. Definitely. So this stiffness-- because when we insert this inner tendon wire. So basically we added one more degree of freedom. 

[00:43:09.93] This will increase the overall kinematics in some direction. So we didn't explain that in great detail because let's assume that if we had tendon wire-- let's assume we have one degree of freedom, but now we add a stimulus. Basically, we have two degrees of freedom. 

[00:43:24.05] So [INAUDIBLE] dexterity, but in some direction we kind of will lack that dexterity. But in some direction we would definitely increase its workspace. So it depends on the location of your target. Sometimes we probably-- without the stiffness modeling, we cannot reach it, but with this stiffness model we do. We can reach that, but in some scenarios we just couldn't reach that target point. 

[00:43:49.37] It depends on your target. Does it makes sense? Thank you. 

[00:43:57.38] PRESENTER: Any other questions? [INAUDIBLE] Yes. 

[00:44:01.41] AUDIENCE: [INAUDIBLE] 

[00:44:04.49] YUE CHEN: Yes. 

[00:44:05.14] AUDIENCE: [INAUDIBLE] OK. 

[00:44:08.22] YUE CHEN: Uh-huh? 

[00:44:09.60] AUDIENCE: [INAUDIBLE] 

[00:44:22.38] YUE CHEN: Yes. 

[00:44:23.50] AUDIENCE: [INAUDIBLE] 

[00:44:27.00] YUE CHEN: Actually, this is our ongoing work. So in our simulation we consider the nerve, consider the DTI. We consider several very critical nerves like the eye nerve and also blood vessels. So what we are trying to do is we want to avoid these critical structures while we are reaching to this too. 

[00:44:45.57] So there will be some optimization work inside the path planning. So we are still working on it right now. Thank you. 

[00:44:55.35] [APPLAUSE]