[CSE 276F Submission] PokeCube-v1 #375
Conversation
| head_to_cube_dist = torch.linalg.norm( | ||
| self.peg_head_pos[..., :2] - self.cube.pose.p[..., :2], axis=1 | ||
| ) | ||
| is_peg_cube_touch = head_to_cube_dist <= self.cube_half_size + 0.005 |
There was a problem hiding this comment.
please rename is_peg_cube_touch to something like is_peg_cube_close. This condition does not check correctly if they are touching. Checking closeness is fine.
There was a problem hiding this comment.
I have modified the variable names containing "touch" incorrectly, thanks for your advice.
| # initialize the goal region | ||
| goal_region_xyz = cube_xyz + torch.tensor([0.05 + self.goal_radius, 0, 0]) | ||
| goal_region_xyz[..., 2] = 1e-3 | ||
| goal_region_q = euler2quat(0, np.pi / 2, 0) |
There was a problem hiding this comment.
can you make the task slightly more difficult by randomizing the quaternion for the cube? This seems too easy. I expect it to be still solvable with a bit of z-axis randomization
There was a problem hiding this comment.
Add z-axis randomization of the cube, rotate in the range [-pi/6, pi/6]
|
Thank you for your submission for the course. This task is nice and sufficiently interesting to be added. However could you make a different peg instead of the default provided magnet/blue red one? Since the peg does not need specific orientation |
The peg has been changed to single-color |
* add new env PokeCube-v1 * add z-axis randomization of cube, modify peg's color, variable name
A simple task where the objective is to poke a red cube with a peg and push it to a target goal position.
Solvable via ppo baseline: