Reinforcement learning of robotic control policies

Researcher:
Prof. Miriam Zacksenhouse | Mechanical Engineering

Categories:

Automation, Mobility and Aerospace | Information and Computer Science

The Technology

Robots are expected to propagate over different terrains. There is a growing need to provide a method for controlling a movement of a robot that is robust, can allow the robot to propagate over terrains with positive and negative slopes, and is cost effective.

A method and a device for controlling a movement and multiple steps of a robot. The method can include multiple iterations of: (i) calculating or receiving a first slope attribute indicative of a slope of a first area of a terrain on which a first leg of the robot steps; (ii) feeding the first slope attribute to a central pattern generator (CPG); and (iii) generating, by the CPG and in response to the slope attribute, at least one control pulse for controlling a torque characteristic of a torque applied by at least one leg of the robot.

Advantages

  • Cost effective
  • Controlling multiple steps of the robot
  • Controlling a movement of a robot
  • The method and devices can be applicable to robots that have: knee/ ankle joints, more than two joints per leg, more than two legs with different configurations, have torso and more.

Applications and Opportunities

  • The methods and devices can be applied to various robotic applications, in particular for companion robots, which have to maneuver in human accessible environments.
arrow Business Development Contacts
Ilia Baskin
Director of Business Development, Engineering