Some custom-designed environments for reinforcement learning based on panda-gym.
Pre-trained agents are provided, some of which are from original environments in panda-gym.
May serve as a tutorial for customizing your own environments.
panda-gym == 2.0.1
stable-baselines3 == 1.5.1a5
sb3-contrib == 1.5.1a5
Files can be found under the "trained_agents" folder.
To enjoy the trained agents, run xxx_show.py.
To re-train an agent, run xxx_train.py.
control-type: joints
The robot is supposed to pass two points successively.
The robot is supposed to pass three points successively.
The robot is supposed to push the object to pass two points successively.
The robot is supposed to push two object to their targets respectively.
The robot learns to push the object when its target point is on the table and pick the object when its target point is in the air.
This is different from the robot trained by dense reward.
The robot is supposed to pick the object and carry it to pass two points successively.
The robot is supposed to reach a point while balancing a ball on its plate.
control-type: joints
The robot is supposed to reach two points successively while balancing a ball on its plate.
control-type: joints
