Organizational Unit:
Undergraduate Research Opportunities Program
Undergraduate Research Opportunities Program
Permanent Link
Research Organization Registry ID
Description
Previous Names
Parent Organization
Parent Organization
Organizational Unit
Includes Organization(s)
ArchiveSpace Name Record
Publication Search Results
Now showing
1 - 5 of 5
-
ItemInvestigating Sim-to-Real Transfer and Multi-Agent Learning in Assistive Gym(Georgia Institute of Technology, 2020-12) Schaffer, Holden C.As the world's population grows older on average and the number of available caregivers decreases, assistive robotics pose an opportunity for older adults or people with disabilities to continue receiving the care that they need. Recent work has shown tremendous progress in using deep reinforcement learning to teach robotic caregivers how to properly assist people in simulation, where robots can learn how to interact with humans in a safe, controlled manner. However, transferring what the robot has learned from simulation to reality continues to pose a challenge for assistive robotics, and a gap in the literature exists in finding techniques to overcome this challenge for this particular domain. The first part of this research uses an assistive simulation framework known as Assistive Gym and its simulated drinking environment to test various approaches to sim-to-real transfer for assistive robotics. The end result of this portion of the research is the identification of a series of baseline steps that are necessary to transfer the Drinking task in Assistive Gym to a physical PR2. Next, the avenues for future works are addressed by investigating a few potential modifications to the drinking task which could be implemented for a more successful transfer of policies. The second part of the research investigates how multi-agent learning could be implemented in Assistive Gym. This section implements multi-agent assistance for the bed-bathing environment, then tests the effectiveness of three different algorithms in order to gauge their effectiveness for solving this new multi-agent task. These algorithms include two variations of single-agent Proximal Policy Optimization modified for multi-agent use as well as Multi-Agent Deep Deterministic Policy Gradient. Finally, future works related to multi-agent assistance are discussed, namely choosing alternate implementations of MADDPG and investigating the dressing environment for its greater potential for cooperation between robots.
-
ItemThe Visual Segmentation of Scene Information and Applications in Predictive Haptics(Georgia Institute of Technology, 2020-05) Srirangam, Bharat V.We as humans take our ability to digest a scene and extract its context knowledge to be for granted. There are several senses involved including but not limited to sight, hearing, and touch. This also includes our ability to combine information from the different senses to enrich our understanding. In the healthcare robotics space, a lot of success has been met with emulating or attempting to emulate these abilities of humans in everyday processes. To be specific, one process, context inference, is a very subconscious but powerful process. When someone picks up a glass cup, they can feel that the cup is made of glass where their hand meets the cup but are still able to infer that the rest of the cup is also glass. This is a powerful example of how humans take local information and use it to derive global context. In this paper, we attempt to emulate this process through a PR2 robot by developing a way to segment a scene for the various objects in the scene. The PR2 can then estimate the material of the different objects using a pre-trained neural network that takes in spectroscopy measurements and pictures of a small patch of each object. This would thereby allow the PR2 to emulate the same ability to abstract its local information to a global context as the material of each object would be determined by a small measurement. We set up a table with an arrangement of objects and test different approaches to segmenting this scene to provide points of interest and measurement to the PR2. Some objects that were used include pots, glasses, mugs, sweaters, and scissors. After testing 3 different approaches to segmentation, a 3D analysis based one was able to sufficiently segment the scenes and provided the PR2 with enough information to make proper measurements and make reasonable estimates. Finally, we demonstrate how a PR2 robot can do all of this and leverage this system to estimate the materials of everyday objects so that it can infer interactions with these objects. From this work, we find that we are one step closer to providing robots with the same advantage that we have to mix partial contextual understandings to make better globally informed decisions.
-
ItemSimulating Assistive Robotics Tasks(Georgia Institute of Technology, 2020-05) Gangaram, Vamsee K.In this thesis, I summarize two published research papers [1][2] to which I contributed as an undergraduate researcher. My contributions to this research primarily consisted of implementing realistic human joint limitations and better cloth visualization in Assistive Gym [1], as well as testing out various capacitive sensor designs for the multidimensional capacitive sensing work [2]. Physics-based simulation offers an opportunity for robots to learn to better provide safe and efficient assistance to people. By training robotic controllers in accurate simulations, we can drastically improve data collection and training times as compared to data collection with real robots and real people. Simulation also provides robots with a safe environment to learn, practice, and make mistakes, without having to put real people at risk. In a previous work Erickson et al. introduced Assistive Gym, a simulation framework based on the PyBullet physics engine to simulate various assistive tasks with robot and human interaction [1]. The six assistive tasks modeled are drinking, eating, itch scratching, dressing, bed bathing, and arm manipulation. We also model various human limitations as well as active human cooperation which results in better learned assistance policies. We include four common assistive robots as options for training in the six environments and show how they can be benchmarked for each assistive task. Another work from Erickson et al. on using multidimensional capacitive sensing for dressing and bathing tasks [2] is summarized, and we describe how this sensor can be modelled in simulation to incorporate into Assistive Gym in the future. Overall, Assistive Gym is shown to be an encouraging framework for training assistive robots in simulation and is open source for the research community to build upon.
-
ItemMaterial Classification with Active Thermography on Multiple Household Objects(Georgia Institute of Technology, 2019-05) Chen, HaofengActive thermography is a technique to inject heat into a target sample and observe the temperature change along time. Such a technique enables a robot to perform material classification with machine learning algorithms and infer material properties of its surroundings. We present a study of material classification on 20 household objects of 5 material classes using active thermography, and analyze factors that impact on material classifiers’ performance on generalizing to heating distances and object instances not present during training. By performing a 20-way classification of the object instances, we show that there is potential for classifiers to generalize to unseen objects made from known material classes. The best-performing algorithm trained on 15 object instances at 5 heating distances (20cm, 25cm, 30cm, 35cm, 40cm) gives an accuracy of 71.7% when generalizing to 5 objects that are not in the training set.
-
ItemTowards Material Classification of Scenes Using Active Thermography(Georgia Institute of Technology, 2019-05) Bai, HaopingBy briefly heating the local environment with a heat lamp and observing what happens with a thermal camera, robots could potentially infer properties of their surroundings. However, this form of active thermography introduces large signal variations compared to traditional active thermography, which has typically been used to characterize small regions of materials in carefully controlled settings. We demonstrate that a data-driven approach with modern machine learning methods can be used to classify material samples over relatively large surface areas and variable distances. We also introduce the use of z-normalization to improve material classification and reduce variation due to distance and heating intensity. Our best performing algorithm achieved an overall accuracy of 77.7% for multi-class classification among 12 materials placed at varying distances (20 cm, 30 cm, and 40 cm). The observations were made for 5 seconds with 1s of heating and 4s of cooling. We also provide a demonstration of performance with a multi-material scene.