Physics Forward Co-Design of Small, Enduring, Multimodal Ground Robots for Reconnaissance

Project #1.45 begins 2026.

Our group is developing MARBLE: a Multimodal Adaptive Robot Ball for Land Exploration. This spherical robot is comprised of a spring-like carbon fiber endoskeleton contained within a compliant, pressurized elastomer shell. MARBLE is inexpensive, controllable, untethered, and capable of active driving (via motorized wheels that rotate the shell) and impulsive jumping (via compression and unlatching of the endoskeleton). MARBLE’s “rubber-like” elastic body stores and recovers impact energy, enabling passive bouncing and rolling that extends endurance on rough terrain. An IMU, a panoramic camera, and a thin, optically clear equatorial band provide horizon-stable imagery.

The goal of this project is to establish a physics-forward co-design framework that unifies the relationships between structure, actuation, and control in small multimodal robots. We will examine how material compliance, internal energy storage, and control policy collectively determine energy efficiency, terrain traversability, and perceptual stability. Using physical MARBLE robots as an experimental testbed and a calibrated physics-based simulator (MARBLE-SIM) as a reduced-order digital model, we will integrate reinforcement-learning–driven design exploration with experimental validation to identify and test low-energy locomotion strategies. The results from physical trials will be used to refine the model and generate successive design iterations—closing the loop between simulation, learning, and hardware. This process will yield a generalizable method for predictive, data-driven co-design of energy-aware autonomous systems at the centimeter scale.

We hypothesize that a physics-informed co-design process linking structural compliance, internal energy storage, and control policy will yield quantifiable gains in energy efficiency and autonomy for small ground robots. Specifically, we expect that:

1. A calibrated, physics-forward simulator capturing elastic impacts, terrain deformation, and internal mass coupling will reproduce key energetic trends observed experimentally.
2. Learning-in-the-loop optimization within this simulator will discover locomotion strategies that minimize energy use while maintaining stability and terrain adaptability.
3. Iteratively validating and refining these strategies in hardware will expose the governing parameters and scaling laws that define energy-aware mobility at the centimeter scale.

#1.45