Deep RL applied to a series-elastic snake robot
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
NatNet
Results
model_offlineA3C
.gitignore
A3C.ipynb
CompliantSnake.py
CompliantSnake0.py
Curves.py
DataLogging.py
GroupLock.py
HebiConstants.py
HebiWrapper.py
Optitrack.py
README.md
Snake.py
SnakeEnvironment.py
actionlist.snake
loadmat.py
runData.snake
snakeReward.py
tools.py
tools0.py

README.md

Deep-SEA-Snake

Reinforcement learning on a series elastic actuated snake robot

Database used for offline learning and videos of the results: here

File list

  • CompliantSnake0.py: State of the art compliant controller with 6 windows - used to generate offline data.
  • CompliantSnake.py: Snake controller running the learned model stored in "model_offlineA3C"
  • Results: Videos and Optitrack logs of trial runs, and scripts to plot the results.
  • A3C.ipnyb: IPython notebook which runs the A3C algorithm with 6 workers on the offline database p_experiments.snake located here

Requirements

  • Numpy
  • Tensorflow
  • matplotlib

**Note: The module "hebiapi" is currently not publicly available, but it is only used for sending and recieving low-level joint information to the robot.

Contact:

Guillaume Sartoretti

William Paivine

Yunfei Shi