Augmenting Neural Networks-Based Model Approximators in Robotic Force-Tracking Tasks

TL;DR

This work proposes a novel control strategy leveraging Neural Networks (NNs) to enhance the force-tracking behavior of a Direct Force Controller (DFC), which accounts for the manipulator's tangential velocity, a critical factor in force exertion, especially during fast motions.

Abstract

As robotics gains popularity, interaction control becomes crucial for ensuring force tracking in manipulator-based tasks. Typically, traditional interaction controllers either require extensive tuning, or demand expert knowledge of the environment, which is often impractical in real-world applications. This work proposes a novel control strategy leveraging Neural Networks (NNs) to enhance the force-tracking behavior of a Direct Force Controller (DFC). Unlike similar previous approaches, it accounts for the manipulator's tangential velocity, a critical factor in force exertion, especially during fast motions. The method employs an ensemble of feedforward NNs to predict contact forces, then exploits the prediction to solve an optimization problem and generate an optimal residual action, which is added to the DFC output and applied to an impedance controller. The proposed Velocity-augmented Artificial intelligence Interaction Controller for Ambiguous Models (VAICAM) is validated in the Gazebo simulator on a Franka Emika Panda robot. Against a vast set of trajectories, VAICAM achieves superior performance compared to two baseline controllers.

Figures (6)

• Figure 1: Simulation setup --- the Panda robot performs a force-tracking task sliding on a wooden table with a spherical-tip end-effector.
• Figure 2: Control architecture. The MA learns the transition function $\bm{\hat{\mathbfcal F}}$ of the dynamical system represented by the impedance-controlled robot interacting with an unknown environment, given data composed of the system states $\bm s$ and control inputs $\bm x_f$, i.e. the DFC action. After training, VAICAM computes the optimal residual action $\bm x_c^*$, aiming at minimizing the force tracking error between $\bm h_r$ and the predicted wrench $\hat{\bm h}_e$.
• Figure 3: MA architecture. The hidden layers approximate the dynamics transition function $\bm s_{k+1} = \mathbfcal F(\bm s_k, \bm x_f)$ by predicting the state variation $\bm\delta \hat{\bm s}$.
• Figure 4: Comparison between the static and the dynamic model approximators, in terms of RMSE, as the EE velocity increases.
• Figure 5: Comparison between the baseline DFC roveda_sensorless_2021, ORACLE petrone_optimized_2025, and VAICAM (ours), in terms of RMSE, as the EE velocity increases.