This environment is part of the tactile regression environments. Refer to the tactile regression environments overview for a general description of these environments.
| Environment ID | ABCCenterOfMass-v0 |
| Dataset | ABC Dataset |
| Prediction Dimensions | 2 |
| Step limit | 32 |
| Sensor rotation | enabled |
| Object pose perturbation | enabled |
In the ABCCenterOfMass environment, the agent's objective is to locate an object from the ABC Dataset positioned randomly on a platform and estimate its precise 2D center of mass projected onto the plate. Unlike the Toolbox task, in which the object shape is known a-priori, in ABCCenterOfMass, the agent faces a completely unknown object in every episode. Hence, this task tests the agent’s ability to both find and thoroughly explore the shape of an unknown object through sequential tactile exploration.
The prediction target is a 2-element np.ndarray containing the 2D coordinates of the object's center of mass in platform coordinates, normalized to the range
import ap_gym
env = ap_gym.make("ABCCenterOfMass-v0")
# Or for the vectorized version with 4 environments:
envs = ap_gym.make_vec("ABCCenterOfMass-v0", num_envs=4)v0: Initial release.
| Environment ID | Description | Preview |
|---|---|---|
| ABCCenterOfMass-Depth-v0 | Uses a depth image instead of rendering tactile images. |  |