mobilePanda slowly collapse. With action and velocity set to zero.

[Quirina300]

The mobile panda slowly collapses even when the actions and velocity are set to zero. The code we used to get this issue:

import gymnasium as gym
from urdfenvs.robots.generic_urdf import GenericUrdfReacher
import numpy as np

def run_mobile_reacher(n_steps=10000, render=False, goal=False, obstacles=True):
    robots = [
        GenericUrdfReacher(urdf="mobilePanda.urdf", mode="vel"),
    ]
    env = gym.make(
        "urdf-env-v0",
        dt=0.01, robots=robots, render=render
    )

    action = np.zeros(env.n())
    vel0 = np.zeros(env.n())
    pos0 = np.array([ 0., 0.,  0.,  0.,  0.,  0., 0., 0.,  1.8675,  0.,  0.02,  0.02])
    # pos0[3] = 0 

    ob = env.reset(pos=pos0, vel=vel0)
    print(f"Initial observation : {ob}")
    history = []
    print(env)
    for i in range(n_steps):
        action[3] = 0
        ob, *_ = env.step(action)
        history.append(ob)
    env.close()
    return history

if __name__ == "__main__":
    run_mobile_reacher(render=True)

[maxspahn]

Hi @Quirina300 ,

Thanks for reporting this issue. I played around with the example and found that the problem is related to numerical inaccuracies. Specifically, pybullet's velocity controller cannot ensure that the target-velocity is reached within one time step. See https://pybullet.org/Bullet/phpBB3/viewtopic.php?t=12688 for a detailed discussion.

One solution to that is to increase the number of sub_steps performed in during one time step. The current setup in urdfenvs is num_sub_steps = 20. https://github.com/maxspahn/gym_envs_urdf/blob/57b82e7ead09941450f8b53200b6b1fd3c9811f7/urdfenvs/urdf_common/urdf_env.py#L85

When this value is increase to 200, or the time step dt=0.01 is decreased to dt=0.001 (the same effect), the problem is solved. However, it is currently not supported to change the value for num_sub_steps from the front-end. I will create a PR on that.