Physics-data hybrid dynamic model of a multi-axis manipulator for sensorless dexterous manipulation and high-performance motion planning

TL;DR

This work develops a physics-data hybrid dynamic model for a 6-DOF manipulator to enable sensorless dexterous manipulation and efficient motion planning. By comparing physics-based EOMs, data-driven models (DNN, LSTM, XGBoost), and their hybrids, it shows that a hybrid approach (H1) achieves the best RMSE with far less training data than pure data-driven methods, while XGBoost-based models can match physics-based performance with larger data sets. A virtual force sensor is built from an external torque observer to infer end-effector forces/torques, validated against a 6-axis sensor in wiping and peg-in-hole tasks, with PMs achieving MAEs on the order of 0.3–0.4 Nm and ~9 N for force components. A PPO-based motion planner leveraging the hybrid model demonstrates ~20% reduction in trajectory time, highlighting the practical value of integrating physics and data-driven corrections for industrial robotic control and planning.

Abstract

We report on the development of an implementable physics-data hybrid dynamic model for an articulated manipulator to plan and operate in various scenarios. Meanwhile, the physics-based and data-driven dynamic models are studied in this research to select the best model for planning. The physics-based model is constructed using the Lagrangian method, and the loss terms include inertia loss, viscous loss, and friction loss. As for the data-driven model, three methods are explored, including DNN, LSTM, and XGBoost. Our modeling results demonstrate that, after comprehensive hyperparameter optimization, the XGBoost architecture outperforms DNN and LSTM in accurately representing manipulator dynamics. The hybrid model with physics-based and data-driven terms has the best performance among all models based on the RMSE criteria, and it only needs about 24k of training data. In addition, we developed a virtual force sensor of a manipulator using the observed external torque derived from the dynamic model and designed a motion planner through the physics-data hybrid dynamic model. The external torque contributes to forces and torque on the end effector, facilitating interaction with the surroundings, while the internal torque governs manipulator motion dynamics and compensates for internal losses. By estimating external torque via the difference between measured joint torque and internal losses, we implement a sensorless control strategy which is demonstrated through a peg-in-hole task. Lastly, a learning-based motion planner based on the hybrid dynamic model assists in planning time-efficient trajectories for the manipulator. This comprehensive approach underscores the efficacy of integrating physics-based and data-driven models for advanced manipulator control and planning in industrial environments.

Figures (16)

• Figure 1: The manipulator TM5-700: (a) The CAD model, (b) the configuration, and (c) the simplified dynamic model.
• Figure 2: Structural diagrams of (a) DNN, (b) LSTM, and (c) XGBoost. The cyan and yellow circles represent the input and output of the model, respectively. In the case of XGBoost, shown in (c), only one of the outputs is the actual output, depending on the result of the final judgment of the input value in the decision tree.
• Figure 3: The inputs/features and outputs/labels of the model using the LSTM method (a) and XGBoost method (b).
• Figure 4: The combination of simplified bounding-box model of the manipulator and hybrid dynamic model of the TM5-700 as a simulator.
• Figure 5: The architecture of the external torque observer.