Robot Perception using reinforcement learning
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
.idea
Documentation
src
.catkin_workspace
.gitignore
1.json
LICENSE
OscilateSensors.json
README.md
USAGE.md
USAGE.md~
test.png

README.md

Robot Perception

Within this repository we are attempting to achieve computational perception via Reinforcement learning and robots.

There are several parts to this and thus, several layers of documentation.

ROS Setup

  • Provides steps required to install ROS, and begin to send command to a dynamixel servo, as well as reading data from the servo.

General value function experiments

  • Describes answering 3 predictive questions about the future using 3 different general value functions.

Horde and Pavlovian Control

  • Presents an architecture that is capable of learning thousands of general value functions in parallel, based of just the sensorimotor information from the servos.
  • Demonstrates how to alter the behavior of the robot based on these predictions
  • Demonstrates how to use error measures that reflect learning progress.

Policy gradient

  • Presents an example of using policy gradient to control the actuator directly be adjusting the policy rather than by adjusting the value function.
  • Demonstrates using Actor Critic methods
  • Demonstrates using continuous actions - by parameterizing the mean and variance by which actions are taken.