Failure-Mode Analysis of A Learned Dexterous Hand Controller

Shadow Dexterous Hand handling a pen shaped object

Completed as part of a graduate course at the University of Toronto | Tensorflow, Python, OpenAI gym.

  • Conducted a failure mode analysis on learned DDPG-based policies used to control a dexterous hand with tactile sensors, in an attempt to understand the utility of tactile information.
  • The experiments were conducted using the OpenAI gym, Mujoco Simulator, and Tensorflow.
  • Video, a full report, and the corresponding code is available here.
Abhinav Grover
Abhinav Grover
Machine Learning Engineer

Machine Learning and Robotics enthusiast, currently building smart robots at Kindred AI (San Francisco).