Emanuel Todorov

University of Washington

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Trends

Author Order

Document Type

Co-Authors

Kendall Lowrey

University of Washington

Aravind Rajeswaran

University of California, Berkeley

Yuval Tassa

DeepMind (United Kingdom)

Vikash Kumar

Sarala Birla University

Tom Erez

DeepMind (United Kingdom)

Krishnamurthy Dvijotham

DeepMind (United Kingdom)

Maryam Fazel

University of Washington

Igor Mordatch

Google (United States)

Svetoslav Kolev

University of Washington

Sergey Levine

University of California, Berkeley

Cooperative Institutions

University of Washington

University of California, San Diego

Google (United States)

University of California, Berkeley

Seattle University

Massachusetts Institute of Technology

Stanford University

Harvard University

University of Washington Applied Physics Laboratory

University of California System

Author Statistics

Papers

Citation

H-Index

i-10 index

Research Field

Virtual Environment Training Improves Motor Performance in Two Patients with Stroke

Neurology Report (1999)

Maureen K. Holden Emanuel Todorov Janet Callahan Emilio Bizzi

Two subjects with hemiplegia were treated using a computer-generated virtual environment (VE) to train upper extremity reach in the impaired limb. Subjects were evaluated pre- and post-VE training using motor recovery and functional ability tests and a real-world test that required reaching to 9 locations in the workspace. During VE training, subjects practiced a virtual task similar to the real task, trying to imitate a virtual teacher's performance. Post-training, reaching errors during real-world performance were reduced by ˜50% (mean across all workspace locations). Both subjects improved in the trained task, indicating transfer of skill from VE to real world performance; reaching also improved in untrained parts of the workspace, indicating some generalization of transfer. Motor recovery and functional scores showed little to no change but one subject acquired the ability to perform several functional tasks (not on the formal test). Our results suggest that VE training holds future promise for stroke rehabilitation.

Work space

Virtual training

Motor Learning

Motor Imagery

10.1097/01253086-199923020-00013

Cite

Citations (121)

STAC: Simultaneous tracking and calibration

Tingfan Wu Yuval Tassa Vikash Kumar Javier R. Movellan Emanuel Todorov

System identification is an essential first step in robotic control. Here we focus on the calibration of kinematic sensors, such as joint angle potentiometers, tendon/actuator extension sensors and motion capture markers, on complex humanoid robots.

Tracking (education)

Identification

10.1109/humanoids.2013.7030016

Cite

Citations (9)

A Minimal Intervention Principle for Coordinated Movement

Neural Information Processing Systems (2002)

Emanuel Todorov Michael I. Jordan

Behavioral goals are achieved reliably and repeatedly with movements rarely reproducible in their detail. Here we offer an explanation: we show that not only are variability and goal achievement compatible, but indeed that allowing variability in redundant dimensions is the optimal control strategy in the face of uncertainty. The optimal feedback control laws for typical motor tasks obey a minimal intervention principle: deviations from the average trajectory are only corrected when they interfere with the task goals. The resulting behavior exhibits task-constrained variability, as well as synergetic coupling among actuators—which is another unexplained empirical phenomenon.

Phenomenon

Motor Control

Source

Cite

Citations (94)

Modeling and identification of pneumatic actuators

Yuval Tassa Tingfan Wu Javier R. Movellan Emanuel Todorov

Pneumatic actuators are mechanically simple and robust, have good energetic properties due to air compressibility, and are relatively cheap. Despite these advantages they are difficult to control - pressure dynamics have typical timescales on the order of 100ms, and this delay can severely cripple simplistic control approaches. The solution is to use a model-based controller with a good model of the pressure dynamics. Here we present a general parametric model of these dynamics based on both a theoretical analysis and an empirical study with a humanoid robot.

Identification

Pneumatic actuator

10.1109/icma.2013.6617958

Cite

Citations (25)

Hierarchical optimal control of a 7-DOF arm model

IEEE Symposium on Adaptive Dynamic Programming and Reinforcement Learning (2009)

Dan Liu Emanuel Todorov

We apply a hierarchical control framework to a realistic arm model. With 7 degrees of freedom and 14 muscles, this arm model has complex nonlinear dynamics operating on 28-dimensional state space and 14-dimensional control space. A high-level controller is designed to capture the main features of the complex high-dimensional plant dynamics but with reduced dimensionality. This allows us to solve the original optimal control problem without running into the curse of dimensionality. We also provide a method to design a low-level controller to generate arm configurations that are consistent with high-level controls and at the same time satisfy biological constraints. To the best of our knowledge, this is the first feedback controller for a detailed 3-D model of a human arm.

10.1109/adprl.2009.4927525

Cite

Citations (28)

MuJoCo HAPTIX: A virtual reality system for hand manipulation

Vikash Kumar Emanuel Todorov

Data-driven methods have lead to advances in multiple fields including robotics. These methods however have had limited impact on dexterous hand manipulation, partly due to lack of rich and physically-consistent dataset as well as technology able to collect them. To fill this gap, we developed a virtual reality system combining real-time motion capture, physics simulation and stereoscopic visualization. The system enables a user wearing a CyberGlove to "reach-in" the simulation, and manipulate virtual objects through contacts with a tele-operated virtual hand. The system is evaluated on a subset of tasks in the Southampton Hand Assessment Procedure - which is a clinically validated test of hand function. The system is also being used by performer teams in the DARPA Hand Proprioception & Touch Interfaces program to develop neural control interfaces in simulation. The software is freely available at www.mujoco.org.

Motion Capture

10.1109/humanoids.2015.7363441

Cite

Citations (97)

Identification and control of a pneumatic robot

Emanuel Todorov Chunyan Hu Alex Simpkins Javier R. Movellan

Pneumatic actuators have a number of advantages over electric motors, including strength-to-weight ratio, tunable compliance at the mechanism level, robustness, as well as price. Their properties are in many ways similar to muscle properties, which further makes them a good choice for bio-inspired robotic designs. However they are considered harder to control, and often avoided. Here we report results on modeling and control of a 2-dof robot, as well as preliminary results on a state-of-the-art 38-dof humanoid. Contrary to popular belief, we found it surprisingly easy to work with these pneumatically-actuated robots and obtained high tracking performance. We were also able to achieve end-effector control of the highly redundant arm of the humanoid. This was done by building parametric models, fitting them to experimental data, designing model-based feedback controllers, and optimizing their performance.

Robustness

Pneumatic artificial muscles

Pneumatic actuator

10.1109/biorob.2010.5627779

Cite

Citations (31)

Control using Recurrent Neural N

Emanuel Todorov

Source

Cite

Citations (0)

Synthesis and stabilization of complex behaviors through online trajectory optimization

2011 IEEE/RSJ International Conference on Intelligent Robots and Systems (2012)

Yuval Tassa Tom Erez Emanuel Todorov

We present an online trajectory optimization method and software platform applicable to complex humanoid robots performing challenging tasks such as getting up from an arbitrary pose on the ground and recovering from large disturbances using dexterous acrobatic maneuvers. The resulting behaviors, illustrated in the attached video, are computed only 7 × slower than real time, on a standard PC. The video also shows results on the acrobot problem, planar swimming and one-legged hopping. These simpler problems can already be solved in real time, without pre-computing anything.

Trajectory Optimization

10.1109/iros.2012.6386025

Cite

Citations (718)

Learning Complex Dexterous Manipulation with Deep Reinforcement Learning and Demonstrations

arXiv (Cornell University) (2017)

Aravind Rajeswaran Vikash Kumar Abhishek Gupta John Schulman Emanuel Todorov

Dexterous multi-fingered hands are extremely versatile and provide a generic way to perform a multitude of tasks in human-centric environments. However, effectively controlling them remains challenging due to their high dimensionality and large number of potential contacts. Deep reinforcement learning (DRL) provides a model-agnostic approach to control complex dynamical systems, but has not been shown to scale to high-dimensional dexterous manipulation. Furthermore, deployment of DRL on physical systems remains challenging due to sample inefficiency. Consequently, the success of DRL in robotics has thus far been limited to simpler manipulators and tasks. In this work, we show that model-free DRL can effectively scale up to complex manipulation tasks with a high-dimensional 24-DoF hand, and solve them from scratch in simulated experiments. Furthermore, with the use of a small number of human demonstrations, the sample complexity can be significantly reduced, which enables learning with sample sizes equivalent to a few hours of robot experience. The use of demonstrations result in policies that exhibit very natural movements and, surprisingly, are also substantially more robust.

Scratch

Sample (material)

Source

Cite

Citations (99)