Co-Design of Rover Wheels and Control using Bayesian Optimization and Rover-Terrain Simulations

Technical Summary: Co-Design of Rover Wheels and Control using Bayesian Optimization and Rover-Terrain Simulations

Problem Statement
While simulation is a cornerstone of robotic system optimization, the high computational cost of modeling deformable terrain has historically restricted its application to full-vehicle studies of off-road autonomous mobility. Traditional approaches relying on the Discrete Element Method (DEM) are often limited to single-wheel tests. This limitation obscures the critical, coupled interactions between wheels, the vehicle body, and the controller, thereby preventing the joint optimization of mechanical design and control strategies. Consequently, the field lacks a scalable framework to co-design rover geometry and control parameters under high-fidelity, full-vehicle closed-loop conditions.

Methodology
To address these limitations, the authors propose a Bayesian optimization framework designed to co-design rover wheel geometry and steering controller parameters. The core of this methodology relies on a continuum-representation model (CRM) for terramechanics, which offers a balance of efficiency and scalability compared to traditional DEM.

The framework operates as follows:

• Simulation Environment: It utilizes high-fidelity, full-vehicle closed-loop simulations on deformable terrain, towing a fixed load across trajectories of varying complexity.
• Optimization Targets: The Bayesian optimizer simultaneously tunes mechanical wheel parameters (radius, width, and grouser features) and control parameters (steering PID gains).
• Objective Function: The system employs a multi-objective formulation that seeks to balance three key performance metrics: traversal speed, tracking error, and energy consumption.
• Comparative Strategy: The study evaluates two distinct design strategies:
  1. Simultaneous Co-Optimization: Optimizing wheel and controller parameters together.
  2. Sequential Approach: Decoupling the mechanical design from the control design.

Key Results
The research demonstrates significant improvements in both computational efficiency and design capability:

• Computational Efficiency: The proposed CRM-based workflow completed 3,000 full-vehicle simulations in five to nine days. In contrast, the authors' previous workflow based on DEM required months to achieve similar scale, highlighting a drastic reduction in computational time.
• Design Trade-offs: The study successfully analyzed the trade-offs between performance and computational cost when comparing simultaneous co-optimization against the sequential approach.
• Hardware Validation: A preliminary hardware study was conducted using the physical rover. The results indicated that the wheel designs optimized in simulation preserved relative performance trends when tested on the physical hardware, suggesting the validity of the simulation-to-reality transfer.

Significance and Claims
The paper claims that scalable, high-fidelity simulation can now enable the practical co-optimization of wheel design and control for off-road vehicles on deformable terrain, removing the reliance on prohibitively expensive DEM studies. By integrating efficient terramechanics modeling with Bayesian optimization, the authors demonstrate that joint mechanical and control design is feasible within a reasonable timeframe. Furthermore, the authors emphasize the reproducibility of their work by releasing the simulation infrastructure, including scripts and models, as open source to support further research in the field.