Dexsensorless: Sensorless Contact Estimation for Dexterous Manipulation

🔗 Project Homepage

Dexsensorless is an open-source framework that estimates contact forces from joint torques using robotic dynamics, enabling sensorless force estimation in dexterous manipulation tasks.

Note: This paper is currently under review. Additional code and resources will be released upon acceptance.

---

🚧 TODO

• Identify base inertia parameters for the robot hand
• Implement momentum-based collision detection
• Expand PyQt UI for easier finger/link selection
• Validate real-world force estimation via collision experiments
• Training code for imitation learning

---

⚙️ Installation

First, install the SymPyBotics

git clone https://github.com/cdsousa/SymPyBotics.git
cd sympybotics
python setup.py install

Then, clone the repository and navigate into the directory:

git clone https://github.com/Prevalenter/DynaForce.git

• Python 3.8 is required.
• If you want to try the interactive simulation demo, make sure to install PyQt5.
• All package versions are listed in utils/python_version.txt.

If you have any questions, please submit an issue.

📖 Project Overview

The repository is organized to support both simulation and real-world experiments for dynamic identification and force estimation.

• data/: Contains robot models (.pkl), sample motion data, and identification results.
• ident/: Core scripts for identification and estimation.
  • ident/real/: Scripts for experiments with the real robot, including multi-algorithm identification (ident_multi_algo.py) and momentum-based force prediction (fore_calibrate/pred_momentum.py).
  • ident/sim/: Scripts for simulation, including an interactive demo with a GUI (sim_interaction_qt.py).
• utils/: Utility functions, including robot model generation and data processing.

The typical workflow involves:

1. Running an identification script from ident/real/ to compute the robot's dynamic parameters from collected data.
2. Using these parameters in either the simulation (ident/sim/) or real-world force estimation scripts.

---

🚀 Getting Started

🔍 Run Identification

cd ident/real/
python ident_single.py

---

🧪 Try in Simulation with UI (Recommended)

We deployed our algorithm in the simulation, allowing the selection of force-applied members through an interactive UI, and printed the detection results.

cd ident/sim/
python sim_interaction_qt.py

---

📡 Momentum-Based Force Estimation (Real World)

cd ident/real/fore_calibrate/
python pred_nacfo.py --K 2.5

---

🙏 Acknowledgements

This project builds upon the work of several amazing open-source libraries, including:

• GX11: For dexterous hand hardware.
• SymPyBotics: For robot dynamic identification.
• LeRobot: For state-of-the-art imitation learning.

We are grateful to the developers and communities behind these projects.

---