cheneating716/diff_rndf

Official Implementation for RNDF: "Differentiable Robot Neural Distance Function for Adaptive Grasp Synthesis on a Unified Robotic Arm-Hand System"

Differentiable Robot Neural Distance Function for Adaptive Grasp Synthesis on a Unified Robotic Arm-Hand System

Yiting Chen, Xiao Gao†, Kunpeng Yao†, Loïc Niederhauser, Yasemin Bekiroglu, Aude Billard

† Equal correspondence.

The code and model will be open sourced upon acceptance.

Robot Neural Distance Function (RNDF)

We construct the Signed Distance Function (SDF) within multi-link articulated robot's joint space. Given arbitrary robot configuration q = [q1,..., qm] and arbitrary query point p=[x, y, z] with its workspace, our RNDF estimates the signed distances d=[d1, ..., dn] between query point p and robot's n links' surface with high precision (with Avg. error of 0.0015m). RNDF supports parallel computation and is differentiable.

Visualizations of ground truth robot mesh and isosurfaces with minimum predicted value=0.001m (solid) and value=0.1m (transparent).

If you find our work useful, please consider citing:

@article{chen2023differentiable,
  title={Differentiable Robot Neural Distance Function for Adaptive Grasp Synthesis on a Unified Robotic Arm-Hand System},
  author={Chen, Yiting and Gao, Xiao and Yao, Kunpeng and Niederhauser, Lo{\"\i}c and Bekiroglu, Yasemin and Billard, Aude},
  journal={arXiv preprint arXiv:2309.16085},
  year={2023}
}

Relative Publications:

Neural-JSDF: Neural joint space implicit signed distance functions for reactive robot manipulator control