A5085247461- Robot Manipulation and Learning
- Robotic Locomotion and Control
- Prosthetics and Rehabilitation Robotics
- Reinforcement Learning in Robotics
- Robotic Mechanisms and Dynamics
- Soft Robotics and Applications
- Modular Robots and Swarm Intelligence
- Human Pose and Action Recognition
- Teaching and Learning Programming
- Advanced Surface Polishing Techniques
- Adhesion, Friction, and Surface Interactions
- 3D Shape Modeling and Analysis
- Neurogenetic and Muscular Disorders Research
- Manufacturing Process and Optimization
- Tactile and Sensory Interactions
- Robotic Path Planning Algorithms
- Model Reduction and Neural Networks
- Computer Graphics and Visualization Techniques
- Hydraulic and Pneumatic Systems
- Music and Audio Processing
- Fashion and Cultural Textiles
- Behavioral and Psychological Studies
- Domain Adaptation and Few-Shot Learning
- Image Processing and 3D Reconstruction
- Speech and Audio Processing
Toyota Research Institute
2018-2024
Toyota Motor Corporation (Switzerland)
2022
MathWorks (United States)
2016
Texas A&M University
2012-2015
In December 2013, 16 teams from around the world gathered at Homestead Speedway near Miami, FL to participate in DARPA Robotics Challenge (DRC) Trials, an aggressive robotics competition partly inspired by aftermath of Fukushima Daiichi reactor incident. While focus DRC Trials is advance for use austere and inhospitable environments, objectives are progress areas supervised autonomy mobile manipulation everyday robotics. NASA's Johnson Space Center led a team comprised numerous partners...
Hybrid zero dynamics (HZD) has emerged as a popular framework for dynamic and underactuated bipedal walking, but significant implementation difficulties when applied to the high degrees of freedom present in humanoid robots. The primary impediment is process gait design-it difficult optimizers converge on viable set virtual constraints defining gait. This paper presents methodology that allows fast reliable generation efficient multi-contact robotic walking gaits through HZD, even presence...
This paper introduces Diffusion Policy, a new way of generating robot behavior by representing robot’s visuomotor policy as conditional denoising diffusion process. We benchmark Policy across 15 different tasks from 4 manipulation benchmarks and find that it consistently outperforms existing state-of-the-art learning methods with an average improvement 46.9%. learns the gradient action-distribution score function iteratively optimizes respect to this field during inference via series...
Hybrid zero dynamics (HZD) has emerged as a popular framework for dynamic walking but significant implementation difficulties when applied to the high degrees of freedom humanoids. The primary impediment is process gait design-it difficult optimizers converge on viable set virtual constraints defining gait. This paper presents methodology that allows fast and reliable generation robotic gaits through HZD framework, even in presence underactuation. Specifically, we describe an optimization...
This paper presents the methodology used to achieve efficient and dynamic walking behaviors on prototype humanoid robotics platform, DURUS. As a means of providing hardware platform capable these behaviors, design DURUS combines highly electromechanical components with "control in loop" leg morphology. Utilizing final DURUS, formal framework for generation gaits which maximizes efficiency by exploiting full body dynamics robot, including interplay between passive active elements, is...
This paper demonstrates the process of utilizing human locomotion data to formally design controllers that yield provably stable robotic walking and experimentally realizing these formal methods achieve dynamically bipedal on NAO robot. Beginning with data, outputs---or functions kinematics---are determined result in a low-dimensional representation locomotion. These same outputs can be considered robot, human-inspired control is used drive robot human. An optimization problem presented...
This paper introduces DextAIRity, an approach to manipulate deformable objects using active airflow.In contrast conventional contact-based quasi-static manipulations, Dex-tAIRity allows the system apply dense forces on out-ofcontact surfaces, expands system's reach range, and provides safe high-speed interactions.These properties are particularly advantageous when manipulating under-actuated with large surface areas or volumes.We demonstrate effectiveness of DextAIRity through two...
This paper tackles the task of goal-conditioned dynamic manipulation deformable objects.This is highly challenging due to its complex dynamics (introduced by object deformation and high-speed action) strict requirements (defined a precise goal specification).To address these challenges, we present Iterative Residual Policy (IRP), general learning framework applicable repeatable tasks with dynamics.IRP learns an implicit policy via delta dynamicsinstead modeling entire dynamical system...
Automating garment manipulation is challenging due to extremely high variability in object configurations. To reduce this intrinsic variation, we introduce the task of "canonicalized-alignment" that simplifies downstream applications by reducing possible This can be considered as "cloth state funnel" manipulates arbitrarily configured clothing items into a predefined deformable configuration (i.e. canonicalization) at an appropriate rigid pose alignment). In end, cloth will result compact...
This paper addresses the problem of controlling underactuated bipedal walking robots in presence actuator torque saturation. The proposed method synthesizes elements Human-Inspired Control (HIC) approach for generating provably-stable controllers, rapidly exponentially stabilizing control Lyapunov functions (RES-CLFs) and standard model predictive (MPC). Specifically, controller uses feedback linearization to construct a linear system describing dynamics outputs. input this is designed be...
Hybrid zero dynamics (HZD) has emerged as a popular framework for the stable control of bipedal robotic gaits, but typically designing gait's virtual constraints is slow and undependable optimization process. To expedite boost reliability HZD gait generation, we borrow methods from trajectory to formulate smoother more linear problem. We present multiple-shooting formulation constraints, combining stability-friendly properties with an optimization-conducive problem formulation. showcase...
We introduce a practical robotics solution for the task of heterogeneous bagging, requiring placement multiple rigid and deformable objects into bag. This is difficult as it features complex interactions between highly under limited observability. To tackle these challenges, we propose robotic system consisting two learned policies: rearrangement policy that learns to place fold in order achieve desirable pre-bagging conditions, lifting infer suitable grasp points bi-manual bag lifting....
This paper presents experimentally realized bipedal robotic walking using ideal torque controllers via a novel approach termed the model resolved motion method (IM-RMM), where system's closed-loop dynamics are integrated forward from actual state of hardware to provide desired positions and velocity commands PD controller. By combining this with gaits generated Human-Inspired Control framework, was on DURUS platform, designed built by SRI, achieved minimal system identification. For...
This paper tackles the task of goal-conditioned dynamic manipulation deformable objects. is highly challenging due to its complex dynamics (introduced by object deformation and high-speed action) strict requirements (defined a precise goal specification). To address these challenges, we present Iterative Residual Policy (IRP), general learning framework applicable repeatable tasks with dynamics. IRP learns an implicit policy via delta dynamics—instead modeling entire dynamical system...
We present Universal Manipulation Interface (UMI) -- a data collection and policy learning framework that allows direct skill transfer from in-the-wild human demonstrations to deployable robot policies. UMI employs hand-held grippers coupled with careful interface design enable portable, low-cost, information-rich for challenging bimanual dynamic manipulation demonstrations. To facilitate learning, incorporates carefully designed inference-time latency matching relative-trajectory action...
Audio signals provide rich information for the robot interaction and object properties through contact. These can surprisingly ease learning of contact-rich manipulation skills, especially when visual alone is ambiguous or incomplete. However, usage audio data in has been constrained to teleoperated demonstrations collected by either attaching a microphone object, which significantly limits its pipelines. In this work, we introduce ManiWAV: an 'ear-in-hand' collection device collect...
Recent studies have made significant progress in addressing dexterous manipulation problems, particularly in-hand object reorientation. However, there are few existing works that explore the potential utilization of developed controllers for downstream tasks. In this study, we focus on constrained food peeling. Food peeling presents various constraints reorientation controller, such as requirement hand to securely hold after We propose a simple system learning controller facilitates...
Compliance plays a crucial role in manipulation, as it balances between the concurrent control of position and force under uncertainties. Yet compliance is often overlooked by today's visuomotor policies that solely focus on control. This paper introduces Adaptive Policy (ACP), novel framework learns to dynamically adjust system both spatially temporally for given manipulation tasks from human demonstrations, improving upon previous approaches rely pre-selected parameters or assume uniform...
This paper introduces Diffusion Policy, a new way of generating robot behavior by representing robot's visuomotor policy as conditional denoising diffusion process. We benchmark Policy across 12 different tasks from 4 manipulation benchmarks and find that it consistently outperforms existing state-of-the-art learning methods with an average improvement 46.9%. learns the gradient action-distribution score function iteratively optimizes respect to this field during inference via series...
We present our findings in the gap between theory and practice of using conditional energy-based models (EBM) as an implicit representation for behavior-cloned policies. also clarify several subtle, potentially confusing, details previous work attempt to help future research this area. point out key differences unconditional EBMs, warn that blindly applying training methods one other could lead undesirable results do not generalize well. Finally, we emphasize importance Maximum Mutual...
This paper tackles the task of goal-conditioned dynamic manipulation deformable objects. is highly challenging due to its complex dynamics (introduced by object deformation and high-speed action) strict requirements (defined a precise goal specification). To address these challenges, we present Iterative Residual Policy (IRP), general learning framework applicable repeatable tasks with dynamics. IRP learns an implicit policy via delta -- instead modeling entire dynamical system inferring...