Probabilistic Latent Variable Modeling for Dynamic Friction Identification and Estimation

Victor Vantilborgh,Sander De Witte,Frederik Ostyn,Tom Lefebvre,Guillaume Crevecoeur

2024-12-20

Abstract:Precise identification of dynamic models in robotics is essential to support control design, friction compensation, output torque estimation, etc. A longstanding challenge remains in the identification of friction models for robotic joints, given the numerous physical phenomena affecting the underlying friction dynamics which result into nonlinear characteristics and hysteresis behaviour in particular. These phenomena proof difficult to be modelled and captured accurately using physical analogies alone. This has motivated researchers to shift from physics-based to data-driven models. Currently, these methods are still limited in their ability to generalize effectively to typical industrial robot deployement, characterized by high- and low-velocity operations and frequent direction reversals. Empirical observations motivate the use of dynamic friction models but these remain particulary challenging to establish. To address the current limitations, we propose to account for unidentified dynamics in the robot joints using latent dynamic states. The friction model may then utilize both the dynamic robot state and additional information encoded in the latent state to evaluate the friction torque. We cast this stochastic and partially unsupervised identification problem as a standard probabilistic representation learning problem. In this work both the friction model and latent state dynamics are parametrized as neural networks and integrated in the conventional lumped parameter dynamic robot model. The complete dynamics model is directly learned from the noisy encoder measurements in the robot joints. We use the Expectation-Maximisation (EM) algorithm to find a Maximum Likelihood Estimate (MLE) of the model parameters. The effectiveness of the proposed method is validated in terms of open-loop prediction accuracy in comparison with baseline methods, using the Kuka KR6 R700 as a test platform.

Robotics,Machine Learning,Systems and Control

What problem does this paper attempt to address?

### What problem does this paper attempt to solve?

This paper aims to solve the difficult problem of dynamic friction identification and estimation in robot joints. Specifically, the authors focus on how to accurately model and identify dynamic friction characteristics in robot systems to support applications such as dynamic simulation, control design, friction compensation, and output torque estimation.

#### Main problem background

1. **Limitations of traditional physical models**:
- Physical models are difficult to capture all the phenomena that affect friction dynamics, such as non - linear characteristics and hysteresis behaviour, which makes it difficult for models based on physical assumptions to accurately describe complex friction phenomena.
- Although dynamic friction models (such as the Lugre model and the GMS model) can partially explain these phenomena, their performance is still limited in actual industrial applications, especially in high - and low - speed operations and frequent direction reversals.

2. **Challenges of data - driven methods**:
- Pure data - driven methods (such as neural networks) can capture complex input - output relationships, but usually require a large amount of data and perform poorly in consistency under different speeds, loads, and direction reversals.
- These methods may also lead to non - Markovian state - spaces and internalize potential state estimates, limiting their generality.

#### Proposed solutions

To solve the above problems, the authors propose the following innovations:

1. **Introducing latent dynamic states**:
- Use latent dynamic states to represent unrecognized dynamic characteristics, enabling the friction model to evaluate the friction torque by combining the dynamic state of the robot and the additional information in the latent state.
- The friction model and the latent state dynamics are both parameterized by neural networks and integrated into the traditional lumped - parameter dynamic robot model.

2. **Probabilistic state - space model (PSSM)**:
- Formulate the problem as a standard probabilistic representation learning problem, and use the expectation - maximization (EM) algorithm and the sequential Monte Carlo (SMC) technique to obtain the maximum - likelihood estimate (MLE) of the model parameters.
- This method directly learns the complete dynamic model from noisy sensor measurement data without pre - processing or handling noise.

3. **Experimental verification**:
- Experimental verification was carried out on the KUKA KR6 R700 industrial robot, and the results show that the proposed method is superior to existing methods in terms of open - loop prediction accuracy.

### Summary

The main goal of this paper is to overcome the limitations of traditional physical models and pure data - driven methods in dynamic friction identification by introducing latent dynamic states and probabilistic state - space models, thereby providing a more accurate and robust dynamic friction modeling method. This method not only improves the prediction accuracy but also enhances the understanding and modeling ability of complex friction behaviors.