Physical Reservoir Computing in Hook-Shaped Rover Wheel Spokes for Real-Time Terrain Identification

TL;DR

This work demonstrates that a rover wheel spoke with a hook-shaped geometry can act as a physical reservoir to separate terrain-induced vibration frequencies for real-time, low-power terrain identification. By placing three piezoelectric sensors at optimally chosen spoke positions (guided by FE analysis and Shannon entropy), the system captures distinct frequency bands that enable effective classification with a lightweight SVM (approximately 90% accuracy). The approach also combines vibration analysis with stress amplification insights and uses Euclidean and Mahalanobis distances to estimate similarity to unknown terrains, improving robustness in unstructured environments. Practically, this method offers a low-power, vibration-driven alternative to vision- or lidar-based terrain sensing, suitable for energy-constrained planetary rovers and autonomous ground vehicles, with potential for adaptive feature selection and sensor-design enhancements in future work.

Abstract

Effective terrain detection in unknown environments is crucial for safe and efficient robotic navigation. Traditional methods often rely on computationally intensive data processing, requiring extensive onboard computational capacity and limiting real-time performance for rovers. This study presents a novel approach that combines physical reservoir computing with piezoelectric sensors embedded in rover wheel spokes for real-time terrain identification. By leveraging wheel dynamics, terrain-induced vibrations are transformed into high-dimensional features for machine learning-based classification. Experimental results show that strategically placing three sensors on the wheel spokes achieves 90$\%$ classification accuracy, which demonstrates the accuracy and feasibility of the proposed method. The experiment results also showed that the system can effectively distinguish known terrains and identify unknown terrains by analyzing their similarity to learned categories. This method provides a robust, low-power framework for real-time terrain classification and roughness estimation in unstructured environments, enhancing rover autonomy and adaptability.

Figures (12)

• Figure 1: The hook shaped spoke in a rover wheel from NASA Jet Propulsion Laboratory. We hypothesized that this shape leads to separation of information in the stresses and their frequencies due to the geometric influence on the bending moment distribution.
• Figure 2: Harmonic response of the entire wheel spoke with 3N act on the end tip of the spoke within the 0-500 Hz range. The vibrational behavior is non-uniform, exhibiting resonance peaks at 90 Hz and 200-300 Hz, which correspond to different deformation modes across the structure.
• Figure 3: (A) Time series plot illustrating the variations in horizontal and vertical force components acting on the spoke over discrete time steps. The larger oscillations reflect wheel rotation forces, while the smaller vibrations indicate terrain fluctuations; (B) Finite element analysis (FEA) results showing the equivalent stress distribution on the spoke under a primarily vertical force. The mounting holes serve as fixed constraints, with peak stress reaching 10.524 MPa at the lower curved region; (C) FEA results showing the equivalent stress on the spoke under a horizontal force. The peak stress shifts to the upper fixed point, reaching 16.707 MPa, indicating significant stress concentration.
• Figure 4: (A) Three measurement positions on the spoke, where stress probes are placed to record stress and deformation responses. (B) Time series plot of stress at the three measurement positions, showing variations over time under dynamic loading. (C) Time series plot of deformation at the three measurement positions, illustrating displacement changes due to applied forces. (D) Time series plot of gain, defined as the ratio of stress to deformation at each position, highlighting differences in structural stiffness and response.
• Figure 5: (A) Probability distribution of deformation at three measurement positions, showing variations in deformation magnitude. (B) Probability distribution of stress at three measurement positions, illustrating stress variability under dynamic loading. (C) Probability distribution of gain (stress-to-deformation ratio) at three positions, highlighting stiffness variations. (D) Comparison of Shannon entropy for stress, deformation, and gain across three positions, quantifying the complexity of each parameter.