Rodrigues Network for Learning Robot Actions

2025-06-03

AI Summary

In robotic action learning, mainstream neural architectures (e.g., MLPs, Transformers) lack inductive biases from rigid-body kinematics, limiting their capacity for accurate motion modeling. To address this, we propose RodriNet—a novel neural network architecture that explicitly embeds analytical kinematic priors. Its core innovation is the differentiable Neural Rodrigues Operator, which encodes the geometric structure of 3D rigid-body rotations directly into network layers, yielding geometrically consistent, differentiable, and structured action representations. RodriNet integrates differentiable kinematic modeling with diffusion-based policy learning and supports single-image 3D hand reconstruction. Experiments demonstrate that RodriNet significantly outperforms baseline models on synthetic motion prediction, while also improving accuracy and cross-scenario generalization in robotic imitation learning and single-image hand reconstruction.

Technology Category

• Intelligent Robots: Learning & Optimization for ROB
• Computer Vision: Visual Reasoning & Symbolic Representations
• Machine Learning: Imitation Learning & Inverse Reinforcement Learning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

Understanding and predicting articulated actions is important in robot learning. However, common architectures such as MLPs and Transformers lack inductive biases that reflect the underlying kinematic structure of articulated systems. To this end, we propose the Neural Rodrigues Operator, a learnable generalization of the classical forward kinematics operation, designed to inject kinematics-aware inductive bias into neural computation. Building on this operator, we design the Rodrigues Network (RodriNet), a novel neural architecture specialized for processing actions. We evaluate the expressivity of our network on two synthetic tasks on kinematic and motion prediction, showing significant improvements compared to standard backbones. We further demonstrate its effectiveness in two realistic applications: (i) imitation learning on robotic benchmarks with the Diffusion Policy, and (ii) single-image 3D hand reconstruction. Our results suggest that integrating structured kinematic priors into the network architecture improves action learning in various domains.

Problem

• Learning robot actions with kinematic-aware neural networks
• Improving articulated action prediction in robotics
• Enhancing imitation learning and 3D reconstruction tasks

Innovation

• Neural Rodrigues Operator for kinematics-aware bias
• Rodrigues Network specialized for action processing
• Integrates kinematic priors into neural architecture