asalbright / Pogo-Stick-Jumping

OpenAI gym environment, testing and evaluation.

Pogo Stick Gym Env

Pogo-Stick Model

System Information

This system is used to research the usefulness of reinforcement learning for defining power conservative control strategies for flexible-legged jumping systems. The details on this system can be found (enter publication link).

How to Use

Either:

1. See Concurrent Design/gym_pogo_stick/README.md
   1. FIXME: Needs updating per concurrent design work
2. See Learn Control/gym_pogo_stick/README.md
3. See Learn Mechanical Parameters/pogo_stick_jumping/README.md
   1. FIXME: DOES NOT EXIT

Env Example Parameters

Parameter	Description
numJumps	2, int: number of jumps the pogo completes to terminate an episode
linear	"Linear", string: type of spring used in environment. "Nonlinear"
trainRobust	False, bool: whether or not the env parameters change during training
epSteps	500, int: number of steps to terminate an episode
evaluating	False, bool: whether or not the position of the actuator is randomly set at reset
rewardType	"Height", string: what the agent is going to learn to accomplish. "Efficiency" "SpecHei" "SpHeEf"
specifiedHeight	0.05, float: the height the agent learns to jump to for rewardType "SpecHei" and "SpHeEf"
captureData	False, bool: whether or not to capture the time series data for environment [Time, Reward, Input, RodPos, RodVel, ActPos, ActVel]
saveDataName	None, string: name of the data when captured, default is "Data_xxx_xxx" where xxx_xxx are date and time stamps
saveDataLocation	None, string or Path: location to save the data, default is "Captured_Data" within current working directory

See random_action.py for example use.