Team WPI‐CMU: Achieving Reliable Humanoid Behavior in the DARPA Robotics Challenge

DeDonato, Mathew; Polido, Felipe; Knoedler, Kevin; Babu, Benzun Pious Wisely; Banerjee, Nandan; Bove, Christoper P.; Cui, Xiongyi; Du, Ruixiang; Franklin, Perry; Graff, Joshua P.; He, Peng; Jaeger, Aaron M.; Li, Lening; Berenson, Dmitry; Gennert, Michael A.; Feng, Siyuan; Liu, Chenggang; Xinjilefu, X; Kim, Joohyung; Atkeson, Christopher G.; Long, Xin; Padır, Taşkın

doi:10.1002/rob.21685

Cited by 38 publications

(23 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…12). After the first few targets (2)(3)(4), RTE is able to make the robot reach each target in around 30s compared to MCTS alone that needs around 50 − 60 s.…”

Section: Simulation Resultsmentioning

confidence: 99%

“…While these robots could certainly have been more robust and some falls prevented, even the best robots will encounter unforeseen situations: hardware failures will always be a possibility, especially with highly complex robots in complex environments [2]. For instance, C. Atkeson et al report that the Atlas robot they used in the DARPA Robotics challenge had a "mean time between failures of hours or, at most, days" [1,3].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Reset-free Trial-and-Error Learning for Robot Damage Recovery

Chatzilygeroudis

Vassiliades

Mouret

2018

Robotics and Autonomous Systems

View full text Add to dashboard Cite

The high probability of hardware failures prevents many advanced robots (e.g., legged robots) from being confidently deployed in real-world situations (e.g., post-disaster rescue). Instead of attempting to diagnose the failures, robots could adapt by trial-and-error in order to be able to complete their tasks. In this situation, damage recovery can be seen as a Reinforcement Learning (RL) problem. However, the best RL algorithms for robotics require the robot and the environment to be reset to an initial state after each episode, that is, the robot is not learning autonomously. In addition, most of the RL methods for robotics do not scale well with complex robots (e.g., walking robots) and either cannot be used at all or take too long to converge to a solution (e.g., hours of learning). In this paper, we introduce a novel learning algorithm called "Reset-free Trial-and-Error" (RTE) that (1) breaks the complexity by pre-generating hundreds of possible behaviors with a dynamics simulator of the intact robot, and (2) allows complex robots to quickly recover from damage while completing their tasks and taking the environment into account. We evaluate our algorithm on a simulated wheeled robot, a simulated six-legged robot, and a real six-legged walking robot that are damaged in several ways (e.g., a missing leg, a shortened leg, faulty motor, etc.) and whose objective is to reach a sequence of targets in an arena. Our experiments show that the robots can recover most of their locomotion abilities in an environment with obstacles, and without any human intervention.Comment: 18 pages, 16 figures, 3 tables, 6 pseudocodes/algorithms, video at https://youtu.be/IqtyHFrb3BU, code at https://github.com/resibots/chatzilygeroudis_2018_rt

show abstract

“…12). After the first few targets (2)(3)(4), RTE is able to make the robot reach each target in around 30s compared to MCTS alone that needs around 50 − 60 s.…”

Section: Simulation Resultsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Reset-free Trial-and-Error Learning for Robot Damage Recovery

Chatzilygeroudis

Vassiliades

Mouret

2018

Robotics and Autonomous Systems

View full text Add to dashboard Cite

show abstract

“…The ATLAS Robot. The system under consideration for diagnostics is Boston Dynamics' Atlas Robot, a humanoid robot that has 30 degrees of freedom, weighs approximately 180 Kg, and is 1.8m tall (DeDonato et. al, 2017).…”

Section: System Definitionmentioning

confidence: 99%

“…al, 2017). Its lower limbs, torso and upper arms are hydraulic joints while the forearms and wrists are electronically powered (DeDonato et. al, 2017).…”

Section: System Definitionmentioning

confidence: 99%

A Systems Approach towards Developing a Diagnostic System for Complex Robots

Krishnan

Bhada

2018

INCOSE International Symp

View full text Add to dashboard Cite

Present day robots are highly complex systems. Most robots perform tasks that are time‐critical in environments where human intervention may not be possible. When a failure occurs in such situations, diagnosing it can be a major challenge. To tackle this, a methodology to develop a diagnostic system using a systems approach is presented in this paper. The use of SysML diagrams as a tool to understand system behavior is highlighted. Finally, this methodology is used to build the diagnostic system of a complex humanoid robot.

show abstract

“…Most of the existing approaches have achieved this goal by relying on teleoperation. [16][17][18][19][20][21][22] Supervisory steering and gas commands are sent to the robot to drive the car in Karumanchi et al 23 ; DeDonato and colleagues 24 propose a hybrid solution, with teleoperated steering and autonomous speed control. The velocity of the car, estimated with stereo cameras, is fed back to a proportional integral (PI) controller, whereas light imaging, detection, and ranging (LIDAR), IMU, and visual odometry data support the operator during the steering procedures.…”

Section: Problem Formulation and Proposed Approachmentioning

confidence: 99%

Autonomous car driving by a humanoid robot

Paolillo

Gergondet

Cherubini

et al. 2017

Journal of Field Robotics

View full text Add to dashboard Cite

Enabling a humanoid robot to drive a car requires the development of a set of basic primitive actions. These include walking to the vehicle, manually controlling its commands (e.g., ignition, gas pedal, and steering) and moving with the whole body to ingress/egress the car. We present a sensor‐based reactive framework for realizing the central part of the complete task, consisting of driving the car along unknown roads. The proposed framework provides three driving strategies by which a human supervisor can teleoperate the car or give the robot full or partial control of the car. A visual servoing scheme uses features of the road image to provide the reference angle for the steering wheel to drive the car at the center of the road. Simultaneously, a Kalman filter merges optical flow and accelerometer measurements to estimate the car linear velocity and correspondingly compute the gas pedal command for driving at a desired speed. The steering wheel and gas pedal reference are sent to the robot control to achieve the driving task with the humanoid. We present results from a driving experience with a real car and the humanoid robot HRP‐2Kai. Part of the framework has been used to perform the driving task at the DARPA Robotics Challenge.

show abstract

Team WPI‐CMU: Achieving Reliable Humanoid Behavior in the DARPA Robotics Challenge

Cited by 38 publications

References 20 publications

Reset-free Trial-and-Error Learning for Robot Damage Recovery

Reset-free Trial-and-Error Learning for Robot Damage Recovery

A Systems Approach towards Developing a Diagnostic System for Complex Robots

Autonomous car driving by a humanoid robot

Contact Info

Product

Resources

About