AcTExplore

Active Tactile Exploration on Unknown Objects

Amir-Hossein Shahidzadeh  •  Seong Jong Yoo  •  Pavan Mantripragada  •  Chahat Deep Singh  •  Cornelia Fermüller  •  Yiannis Aloimonos
IEEE International Conference on Robotics and Automation (ICRA) 2024

Website  •  Paper  •  YouTube

TL;DR: We explore and reconstruct the object's surface with robotic fingers equipped with tactile sensor

Content

This package contains several components:

Installation:

• Clone repo

• python==3.9

• Install pytorch >= 2.2.1

• pip install -r requirements.txt

• For Training

  • Generate 'Training/Logs' folder
  • Set configuration at conf/RL.yaml
    • state
      • input_type: TTA, TTS, depth
    • reward
      • type: TM, AM, AMB
  • Execute python PPO.py ....

• For testing

  • Generate 'outputs' folder
  • Set configuration at conf/test.yaml
    • RL
      • algorithm: PPO, RecurrentPPO
      • pretrain_model_path: 'best/model/path'
    • state
      • input_type: TTA, TTS, depth (has to match with pretrain_model settings)
    • reward
      • type: TM, AM, AMB (has to match with pretrain_model settings)
    • environment
      • object
        • object_name: test object name
        • urdf_path: ["objects/ycb/object/model.urdf"]
  • Execute python test.py

Exploration in Tacto

Reconstruction

Bibtex

@misc{shahidzadeh2023actexplore,
      title={AcTExplore: Active Tactile Exploration on Unknown Objects}, 
      author={Amir-Hossein Shahidzadeh and Seong Jong Yoo and Pavan Mantripragada and Chahat Deep Singh and Cornelia Fermüller and Yiannis Aloimonos},
      year={2023},
      eprint={2310.08745},
      archivePrefix={arXiv},
      primaryClass={cs.RO}
}