Research Analyzer
← Back ICRA 2026

Differentiable Contact Dynamics for Stable Object Placement under Geometric Uncertainties

Linfeng Li, Gang Yang, Lin Shao, David Hsu

PDF

AI summary

Key figure (auto-extracted from paper)
A novel gradient-based method aligns simulated and measured force-torque data to estimate unknown object and environment geometries, enabling robust stable placement under uncertainty.
Differentiable simulation Contact dynamics Geometric uncertainty Force-torque sensing Stable placement Robot manipulation

Problem

Stable object placement is highly sensitive to geometric uncertainties, yet existing differentiable simulators cannot estimate these uncertainties because they lack gradients of contact forces with respect to geometric parameters.

Approach

The authors derive and implement gradients of contact wrenches with respect to geometric parameters in a differentiable simulator, using gradient descent to refine geometric estimates from force-torque feedback while maintaining a multi-particle belief to mitigate initialization sensitivity.

Key results

  • Derivation and implementation of contact force gradients with respect to geometric parameters
  • Integration of geometric differentiability into the Jade rigid-body simulator
  • Successful stable placement of objects under in-hand pose, shape, and environment uncertainties on a Franka robot arm
  • Superior placement accuracy and robustness compared to trigger-based and particle filter baselines

Why it matters

Provides a general, model-based framework for robots to perceive and adapt to unknown geometries during contact-rich tasks, advancing reliable autonomous manipulation in unstructured environments.

Abstract

From serving a cup of coffee to positioning mechan- ical parts during assembly, stable object placement is a crucial skill for future robots. It becomes particularly challenging under geometric uncertainties, e.g., when the object pose or shape is not known accurately. This work leverages a differentiable simulation model of contact dynamics to tackle this challenge. We derive a novel gradient that relates force-torque sensor readings to geometric uncertainties, thus enabling uncertainty estimation by minimizing discrepancies between sensor data and model predictions via gradient descent. Gradient-based methods are sensitive to initialization. To mitigate this effect, we maintain a belief over multiple estimates and choose the robot action based on the current belief at each timestep. In experiments on a Franka robot arm, our method achieved promising results on multiple objects under various geometric uncertainties, including the in-hand pose uncertainty of a grasped object, the object shape uncertainty, and the environment uncertainty.

Index terms

Manipulation Planning Model Learning for Control Contact Modeling

Related papers