Implementation:Isaac sim IsaacGymEnvs AllegroKukaRegrasping

Knowledge Sources	IsaacGymEnvs
Domains	Robotic_Manipulation, Dexterous_Grasping
Last Updated	2026-02-15 11:00 GMT

Overview

AllegroKukaRegrasping is a single-arm regrasping task environment where a Kuka arm with an Allegro hand must pick up an object, transfer it between fingers, and regrasp it at a target position.

Description

The AllegroKukaRegrasping class extends AllegroKukaBase to implement a regrasping task in the IsaacGym simulation environment. In this task, the robotic system consisting of a Kuka arm and an Allegro dexterous hand must pick up an object from a surface, manipulate it within the hand by transferring it between fingers, and place or hold it at a randomly sampled target position within a defined target volume.

The class loads a goal ball asset (with gravity disabled) that serves as a visual indicator of the target position. During environment resets, the target position is randomized within a configurable volume, and the object is reset to its initial pose on the table. The task uses a single object keypoint at the origin since object orientation is not relevant for regrasping -- only positional accuracy matters.

A curriculum learning strategy is employed via tolerance_curriculum to progressively decrease the success tolerance threshold as the agent improves. The _true_objective method computes a Population-Based Training (PBT) objective that prioritizes tolerance improvements over raw success counts, ensuring that agents are rewarded for achieving tighter tolerances before optimizing for total successes.

Usage

Use this class when training a reinforcement learning policy for single-arm dexterous regrasping tasks. It is suitable for scenarios where the agent must learn to pick up objects and reposition them within the hand, without concern for the object's rotational orientation. Register it as a task in the IsaacGymEnvs task registry and configure it via Hydra configuration files.

Code Reference

Source Location

• Repository: IsaacGymEnvs
• File: isaacgymenvs/tasks/allegro_kuka/allegro_kuka_regrasping.py
• Lines: 42-123

Signature

class AllegroKukaRegrasping(AllegroKukaBase):
    def __init__(self, cfg, rl_device, sim_device, graphics_device_id, headless, virtual_screen_capture, force_render):
        ...

    def _object_keypoint_offsets(self):
        ...

    def _load_additional_assets(self, object_asset_root, arm_pose):
        ...

    def _create_additional_objects(self, env_ptr, env_idx, object_asset_idx):
        ...

    def _after_envs_created(self):
        ...

    def _reset_target(self, env_ids: Tensor) -> None:
        ...

    def _extra_object_indices(self, env_ids: Tensor) -> List[Tensor]:
        ...

    def compute_kuka_reward(self) -> Tuple[Tensor, Tensor]:
        ...

    def _true_objective(self) -> Tensor:
        ...

    def _extra_curriculum(self):
        ...

Import

from isaacgymenvs.tasks.allegro_kuka.allegro_kuka_regrasping import AllegroKukaRegrasping

I/O Contract

Inputs

Name	Type	Required	Description
cfg	dict	Yes	Hydra configuration dictionary containing environment parameters (target volume, tolerances, curriculum settings)
rl_device	str	Yes	Device string for the RL algorithm (e.g., "cuda:0")
sim_device	str	Yes	Device string for the simulation (e.g., "cuda:0")
graphics_device_id	int	Yes	GPU device ID for rendering
headless	bool	Yes	Whether to run without a display
virtual_screen_capture	bool	Yes	Whether to capture virtual screen output
force_render	bool	Yes	Whether to force rendering even in headless mode

Outputs

Name	Type	Description
rew_buf	Tensor	Per-environment reward buffer computed by compute_kuka_reward
is_success	Tensor	Boolean tensor indicating which environments achieved the goal
true_objective	Tensor	PBT objective combining tolerance progress and success count

Usage Examples

# Register and instantiate the AllegroKukaRegrasping task via IsaacGymEnvs
from isaacgymenvs.tasks.allegro_kuka.allegro_kuka_regrasping import AllegroKukaRegrasping

# Typically created through the task registry with Hydra config
env = AllegroKukaRegrasping(
    cfg=task_cfg,
    rl_device="cuda:0",
    sim_device="cuda:0",
    graphics_device_id=0,
    headless=True,
    virtual_screen_capture=False,
    force_render=False,
)

# Run a step in the environment
obs = env.reset()
actions = policy(obs)
obs, rewards, dones, info = env.step(actions)