Actuation-aware co-design for energy-efficient wall-climbing

Abstract

Strict adhesion, friction, and actuator constraints must be met by wall-climbing robots moving on vertical surfaces while maintaining energy efficiency. Robot morphology, actuation, and control are closely related in these systems; however they are normally designed in a sequential manner, which can lead to low performance and increased energy consumption. In this thesis, a physics-based simulation environment is used to simultaneously optimize mechanical design parameters and control policies in an actuationaware co-design framework for energy-efficient wall climbing. Contact dynamics, adhesion forces, friction constraints, and actuator limits are explicitly represented in a MuJoCo model of a 2 DOF wall-climbing robot. The low-level controller used to produce stable periodic climbing gaits while enforcing physical constraints is a nonlinear Model Predictive Controller (NMPC). In order to automatically adjust the internal cost weights of the NMPC, learning-based and evolutionary optimization techniques are used as outer-loop optimizers. Proximal Policy Optimization (PPO) and Covariance Matrix Adaptation Evolution Strategy (CMA-ES) are assessed and compared. In order to reduce energy consumption per vertical meter while preserving climbing feasibility and stability, a unified co-design process builds upon the optimized controller by jointly optimizing leg length, hip joint range, and motor gear ratio. According to simulation results, the suggested co-design approach outperforms control-only optimization and manually adjusted baselines in terms of climbing performance and energy consumption. For dependable and effective wall-climbing locomotion, integrated morphology–actuation–control optimization is crucial, as demonstrated by the optimized system's robust behaviour under various surface friction conditions.