Data-driven Interpretable Hybrid Robot Dynamics

2025-12-10

AI Summary

To address the poor interpretability and limited generalizability of black-box models in robotic dynamics modeling, this paper proposes an interpretable hybrid modeling paradigm that integrates analytical models with data-driven residual terms: leveraging rigid-body dynamics as a physical prior, it learns closed-form, physically meaningful residual torque expressions directly in joint space. Innovatively, we are the first to combine symbolic regression with Sparse Identification of Nonlinear Dynamics (SINDy) to automatically discover compact, interpretable analytical residual formulas. In Franka Emika Panda simulations, our method achieves significantly lower relative error than neural networks, with superior accuracy and generalization. On the real-world Barrett WAM platform, it effectively mitigates overfitting and uncovers novel, physically consistent terms—extending the original model—thereby achieving a balanced trade-off among accuracy, interpretability, and physical consistency.

Technology Category

• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Reasoning under Uncertainty: Relational Probabilistic Models
• Machine Learning: Neuro-Symbolic Learning

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• 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

Abstract

We study data-driven identification of interpretable hybrid robot dynamics, where an analytical rigid-body dynamics model is complemented by a learned residual torque term. Using symbolic regression and sparse identification of nonlinear dynamics (SINDy), we recover compact closed-form expressions for this residual from joint-space data. In simulation on a 7-DoF Franka arm with known dynamics, these interpretable models accurately recover inertial, Coriolis, gravity, and viscous effects with very small relative error and outperform neural-network baselines in both accuracy and generalization. On real data from a 7-DoF WAM arm, symbolic-regression residuals generalize substantially better than SINDy and neural networks, which tend to overfit, and suggest candidate new closed-form formulations that extend the nominal dynamics model for this robot. Overall, the results indicate that interpretable residual dynamics models provide compact, accurate, and physically meaningful alternatives to black-box function approximators for torque prediction.

Problem

• Identifies interpretable hybrid robot dynamics using data-driven methods
• Recovers compact closed-form expressions for residual torque from joint-space data
• Provides accurate and generalizable alternatives to black-box torque predictors

Innovation

• Combines analytical rigid-body dynamics with learned residual torque
• Uses symbolic regression and SINDy for compact closed-form expressions
• Provides interpretable, accurate models that outperform neural networks