DDBot: Differentiable Physics-based Digging Robot for Unknown Granular Materials

TL;DR

DDBot addresses the challenging problem of high-precision digging in unknown granular materials by coupling a differentiable MLS-MPM simulator with SVK elasticity and DP plasticity to enable gradient-based optimization. A parameterized digging skill reduces the action space to a small set of learnable parameters, while demonstration priors and a line-search-based optimization ensure robust convergence. The system demonstrates accurate system identification for sand and soil and achieves zero-shot real-world digging with millimeter-scale precision, outperforming trajectory-based optimization and RL baselines. This work highlights the viability of differentiable physics for granular tasks and points to future work on multi-skill planning and 3D scene reconstruction to broaden applicability.

Abstract

Automating the manipulation of granular materials poses significant challenges due to complex contact dynamics, unpredictable material properties, and intricate system states. Existing approaches often fail to achieve efficiency and accuracy in such tasks. To fill the research gap, this paper studies the small-scale and high-precision granular material digging task with unknown physical properties. A new framework, named differentiable digging robot (DDBot), is proposed to manipulate granular materials, including sand and soil. Specifically, we equip DDBot with a differentiable physics-based simulator, tailored for granular material manipulation, powered by GPU-accelerated parallel computing and automatic differentiation. DDBot can perform efficient differentiable system identification and high-precision digging skill optimisation for unknown granular materials, which is enabled by a differentiable skill-to-action mapping, a task-oriented demonstration method, gradient clipping and line search-based gradient descent. Experimental results show that DDBot can efficiently (converge within 5 to 20 minutes) identify unknown granular material dynamics and optimise digging skills, with high-precision results in zero-shot real-world deployments, highlighting its practicality. Benchmark results against state-of-the-art baselines also confirm the robustness and efficiency of DDBot in such digging tasks.

Figures (15)

• Figure 1: The DDBot (Differentiable Digging Robot) system overview. DDBot has five modules. 1) A robot platform (orange) that performs real-world manipulation and collects real-world surface point clouds of the granular materials. The five small images visualise a real-world digging trajectory that matches the trajectory optimised in simulation shown in the green box. 2) A differentiable physics simulator (green) that simulates the digging skill trajectory given the physics and skill parameters. 3) A differentiable loss function (pink) that compares the geometrical differences between the simulated and real manipulation results. 4) A gradient-based optimisation algorithm (yellow) that updates the physics or skill parameters. 5) The Robot Operating System (ROS) communication protocol and the MoveIt! motion planning library Gorner2019Moveit that generates a velocity trajectory from the skill parameters to control the real robot.
• Figure 2: Visualising the digging skill modelled by Eqs. \ref{['eq:d1']} to \ref{['eq:delta_d4']}. Orange dashed line: soil or sand.
• Figure 3: Real manipulation platform
• Figure 4: Digging task goal visualisation. PCD: point cloud.
• Figure 5: Loss landscapes and their gradient distributions.