Skip to content

mrugacz95/octopus-arm-rl

Repository files navigation

Octopus Arm RL

Project demonstrates powers of Reinforcement Learning on Octopus Arm model.

Agent's target is to touch red dot using 3 types of muscles of arm.

octopus_arm

Usage

Environment is implemented in java and communcation with python is provided with sockets.

$ python agent_handler.py --help
usage: agent_handler.py [-h] [--p PORT] [--h HOST] [--t TEST_DIR]
                        [--e EPISODES] [--no_learning] [--plot_score]
                        {internal,external,external_gui}

Example:

python agent_handler.py externa_gui --no_learning --t test_hard

Tests

For leaning procces test has been divided on dirs by the level of difficulty.

  • easy
  • medium
  • hard
  • nightmare

Results

With RL our agent was able to solve all tests and achieve average score 8.35935.

Plot below shows score depending on starting angle of arm:

35398984_1674484212648368_6107770346218192896_n

More information about implementation (PL)

Report

Bibliography

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages