Reasoning and Planning

ActPerMoMa

ActPerMoMa chooses informative base and camera motions so a mobile manipulator can improve object perception before acting.

Tool Introduction

Core parameters, trigger timing, and visual before/after demo references.

Short Explanation

Use ActPerMoMa when a robot should actively move to reduce uncertainty before grasping or manipulating an object.

InputScene belief + robot state + manipulation target

OutputActive perception or manipulation action

Trigger TimingTriggered on demand after the required input files and configuration are prepared.

RuntimePython / Isaac Sim research code

BeforeScene belief + robot state + manipulation target

Prepare the scene, image, video, sensor stream, prompt, or configuration expected by the original project.

AfterActive perception or manipulation action

Read the produced visualization, prediction, map, trajectory, mask, grasp pose, or other documented artifact.

Parameters And Output

Readable controls and the meaning of each returned artifact.

Parameter Explanation

scene_statepath

Current perception state or simulator observation.

target_objecttext

Object or goal that drives the active perception objective.

Output Explanation

viewpoint_action

Motion selected to improve perception of the target.

manipulation_action

Action proposed after the perception state is sufficiently informative.

How To Use

Official resources, deployment steps, academic context, citation, and source-reported benchmark numbers.

Resources

GitHubhttps://github.com/pearl-robot-lab/ActPerMoMa Code Downloadhttps://github.com/pearl-robot-lab/ActPerMoMa/archive/refs/heads/main.zip Paperhttps://arxiv.org/abs/2310.00433

Deployment Notes

Set up the official simulation environment and assets.
Provide the robot state, scene observation, and manipulation target.
Run the active perception policy before issuing final manipulation actions.

Relative Path Example

# Follow the official ActPerMoMa environment setup, then run the provided policy/evaluation scripts in Isaac Sim.

Expected Result Shape

{
  "tool": "actpermoma",
  "status": "ok",
  "results": [
    {
      "label": "Active perception for mobile manipulation",
      "score": 0.87,
      "output": "Active perception or manipulation action"
    }
  ],
  "timing": {
    "runtime": "The paper reports path/view statistics such as dtotal and vtotal rather than a single wall-clock latency.",
    "device": "documented in source benchmark when available"
  },
  "artifacts": {
    "visualization": "tools/actpermoma/runs/visualization.png",
    "raw_predictions": "tools/actpermoma/runs/predictions.json"
  }
}

Paper figure

Academic Info

Paper identity and contribution summary.

TitleActPerMoMa: Active Perception for Mobile Manipulation

AuthorsAdd authors

VenueCoRL 2023

ContributionCouples active perception with mobile manipulation so the robot can move to better viewpoints before executing a task.

Citation

@misc{actpermoma2023,
  title={ActPerMoMa: Active Perception for Mobile Manipulation},
  author={Author},
  year={2023},
  note={CoRL 2023},
  url={https://arxiv.org/abs/2310.00433}
}

Benchmark

Only compact, source-reported numbers are shown here.

Dataset	Metric	Value	Runtime	Source
Simple scenes, 500 episodes	Success / abort / grasp-failure rate	95.4% / 1.4% / 3.2%	dtotal 3.59+/-1.69 m; vtotal 12.67+/-5.39	Official CoRL 2023 paper, Table III
Complex scenes, 500 episodes	Success / abort / grasp-failure rate	92.6% / 2.2% / 5.2%	dtotal 4.57+/-2.51 m; vtotal 16.20+/-8.65	Official CoRL 2023 paper, Table III
Complex scenes, hard grasps, 500 episodes	Success / abort / grasp-failure rate	61.8% / 29.6% / 8.6%	dtotal 7.19+/-4.17 m; vtotal 24.81+/-13.24	Official CoRL 2023 paper, Table III

Artifacts

Official CoRL 2023 paper, Table III, Isaac Sim setup, and active-perception policy evaluation scripts.

Demo Images

Visual references from the original tool. Click any image to inspect the original size.