Permanent URI for this collection
1 - 5 of 6665
ItemDynamic Legged LocoManipulation: Balancing Reinforcement Learning with Model-Based Control( 2023-04-12)Model-based control methods such as control Lyapunov and control barrier functions can provide formal guarantees of stability and safety for dynamic legged locomotion, given a precise model of the system. In contrast, learning-based approaches such as reinforcement learning have demonstrated remarkable robustness and adaptability to model uncertainty in achieving quadrupedal locomotion. However, reinforcement learning based policies lack formal guarantees, which is a known limitation. In this presentation, I will demonstrate that simple techniques from nonlinear control theory can be employed to establish formal stability guarantees for reinforcement learning policies. Moreover, I will illustrate the potential of reinforcement learning for more complex bipedal and humanoid robots, as well as for loco-manipulation tasks that entail both locomotion and manipulation. This brings up an intriguing question: Is reinforcement learning alone sufficient for achieving optimal results in dynamic legged locomotion, or is there still a need for model-based control methods?
ItemAn engineer’s perspective on antiferromagnetic spintronics(Georgia Institute of Technology, 2023-03-07)Antiferromagnets (AFM) materials have ordered spin moments that alternate between individual atomic sites, which gives them a vanishing macroscopic magnetic signature and picosecond intrinsic timescale. In his 1970 Nobel Lecture, Louis Néel claimed that antiferromagnets are “extremely interesting from theoretical standpoint, but do not seem to have any applications.” Traditionally, AFM materials have played a secondary role to ferromagnets, which are used as active elements in commercial spintronic devices like magnetic sensors and non-volatile magnetic memory. However, it was recently suggested that spin transfer torque could in principle be used to manipulate the magnetic order in AFMs, leading to either stable AFM order precessions for their use as high-frequency oscillators, or switching of the AFM order for their use as magnetic memories. My presentation will focus on recent theoretical and experimental developments in the field of spintronic devices using antiferromagnets as their active elements. I will specifically talk about two unique AFM materials, Cr2O3, a single-phase magnetoelectric material that can be manipulated solely with electric fields and the Weyl semi-metal Mn3Sn in which spin torque can induce chiral spin rotations. Cr2O3-based ferromagnet-free random access memory has been experimentally demonstrated, while in the case of Mn3Sn, spin torque driven dynamics were found to induce chiral oscillations, from the megahertz to the terahertz frequency range. These materials can overcome the central challenge of manipulating and reading the AFM’s order parameter via microelectronics compatible circuitry, thus allowing us to develop antiferromagnetic spintronics along a similar route as ferromagnetic spintronics. I will conclude my talk by summarizing the limits, challenges, and opportunities of AFM spintronics for future technologies such as high-density, secure nonvolatile memory, compact narrowband terahertz sources, and spike generators.
ItemPerception in Action: Neural Circuits for Active Auditory and Tactile Decision-making( 2023-04-17)How do we explore and learn about our world? In nature, animals do not passively await stimuli, as they typically must do in the laboratory. Instead, they actively seek out sensory information—for instance, to find food or shelter. I will first present a summary of my postdoctoral work, which showed how mice recognize different shapes using their whiskers. We used a new method called behavioral decoding to show what sensorimotor strategies mice used to recognize shapes, and we identified an efficient formatting for those strategies in somatosensory cortex. Next, I will present new work from my own lab. We have developed an active sound-seeking task for mice, in which they use head and body movements to find sound sources. We hypothesize that sensory and motor brain regions exchange predictive signals to compute how best to move the body to localize the sound. In future work we plan to identify how central sensorimotor plasticity enables resilience to sensory loss, with the ultimate goal of rationally engineering neural interventions to restore healthy sensorimotor function.
ItemCombining Transcranial Brain Stimulation with Neuroimaging for State-dependent Stimulation and Causal Network Interrogation( 2023-03-27)Functional neuroimaging and electrophysiological techniques, such as functional magnetic resonance imaging (fMRI) as well as electro- and magnetoencephalography (EEG/MEG), serve well to study spontaneous or task-related neuronal activity as correlates of specific cognitive functions in the human brain. However, to infer causality of brain activation for cognition, the former must be manipulated experimentally. This is possible in healthy humans with the help of non-invasive brain stimulation (NIBS) techniques, such as transcranial magnetic stimulation (TMS), transcranial electric stimulation (tES), and since recently also transcranial ultrasound stimulation (TUS). Importantly, NIBS can also be combined with fMRI as well as EEG/MEG, either concurrently (online) or consecutively (offline). Online approaches, assessing the immediate neural response to stimulation, can be used to (i) quantify neuronal network properties such as excitation, inhibition, or connectivity, (ii) interfere with ongoing spontaneous or task-related activity and thus affect behavioral performance, or (iii) modulate the level and timing of neuronal activity, e.g., trying to mimic neuronal oscillations in behaviorally relevant manner. In contrast, offline approaches can be utilized to either (iv) inhibit or (v) facilitate local neuronal excitability via the induction of synaptic plasticity, assessing its subsequent effects on neuronal activity and behavior. In this talk, I will discuss the different approaches and challenges with respect to their combination with fMRI and EEG, in particular concurrent TMS-fMRI and TMS-EEG, and highlight their potential as well as the caveats for inferring causality from NIBS studies in cognitive neuroscience. I will also introduce the novel approach of brain state-dependent brain stimulation, which allows to control NIBS in real-time based on the online assessment of specific oscillatory states, providing a unique opportunity to causally interact with ongoing neuronal oscillations to study its role in information processing and synaptic plasticity.
ItemReconnecting the Hand and Arm to the Brain (ReHAB): Bi-directional Neuroprostheses for Sensorimotor Functional Restoration( 2023-02-14)Cortically controlled neuroprostheses have long been posited as the “holy grail” for intracortical brain-machine interfaces (BMIs). The efficacy of BMIs has advanced to the point where a small number of laboratories around the US now run human clinical trials with people with chronic paralysis. As part of the ReHAB Clinical Trial, my Laboratory for Intelligent Machine-Brain Systems (LIMBS) investigates using BMIs to control Functional Electrical Stimulation (FES) systems for restoring reach-to-grasp movements to persons with chronic high cervical spinal cord injury. This lecture will discuss several of our clinical, technological, and scientific advances towards developing a bi-directional BMI controlled FES arm neuroprosthesis for restoring motor and somatosensory function. The highlight of this lecture will be the demonstration of a current ReHAB participant, an individual with chronic tetraplegia, eight years post-injury using a multi-nodal BMI with multi-contact FES nerve cuff electrodes to volitionally and independently perform functional tasks, such as self-feeding and shaking hands, and discerning somatosensory feedback through intracortical microstimulation (ICMS). This lecture will also discuss use of human BMI systems as a platform for interrogating fundamental questions of human sensorimotor control, including understanding underlying mechanisms of motor performance and learning, and internal representations of kinetic, kinematic, and somatosensory parameters. Finally, this lecture will discuss steps towards clinical translation of viable FES+BMI neuroprosthetic systems for potential at-home use.