Planning and Control for Hybrid Locomotion of Wheeled-Legged Robots

Abstract

This dissertation describes an optimization-based framework to perform complex and dynamic locomotion strategies for robots with legs and wheels. The proposed method allows to perform novel maneuvers, which exploit the wheeled-legged robot's full capabilities over challenging obstacles. By combining innovative techniques in motion control and planning, this work reveals the full potential of wheeled-legged robots and their superiority compared to their legged counterparts. The work in this thesis is published in two conference proceedings and three journal articles. The research community in legged robotics focuses on bio-inspired robots, although there are some human inventions that nature could not recreate. One of the most significant examples is the wheel that has made our transportation system more efficient and faster, especially in urban environments. Inspired by this human-made evolution, we developed the wheeled-legged robot ANYmal with non-steerable wheels attached to its legs, allowing the robot to be efficient on flat as well as agile on challenging terrain. This novel platform, with powered wheels, achieves a speed of 4 m/s on flat terrain, overcomes challenging obstacles with 1.5 m/s, and reduces the cost of transport by 83 % compared to legged systems. The superiority in speed and efficiency is further verified through skating motions with passive wheels, reducing the energetic cost by 80 % compared to their legged versions. This enhancement, however, comes at the cost of increased complexity due to additional degrees of freedom at the end-effector, which empower motions along the rolling direction while in contact. Furthermore, the torsos' movement results from contact forces at the wheels following high-dimensional and nonlinear physical laws. The missing examples in nature make designing templates that capture the underlying locomotion principles cumbersome, making the hybrid locomotion problem challenging. In this thesis, we focus on novel motion control and planning frameworks overcoming these challenges. The motion controller relies on the robot's full rigid body dynamics and can track whole-body motion references. To this end, we present a hierarchical whole-body controller that computes optimal generalized accelerations and contact forces by solving a sequence of prioritized tasks, including the nonholonomic rolling constraint. In contrast to related robots, all joints, including the wheels, are torque controlled, allowing an optimization that includes the system dynamics and the adaptation to the terrain. The dissertation's main contributions stem from locomotion planning algorithms that rely on TO and MPC algorithms optimizing the robot's whole-body trajectory over a receding horizon. By breaking down the optimization problem into a wheel and base TO, locomotion planning for high-dimensional wheeled-legged robots becomes more tractable. It can be solved in real-time on-board in an MPC fashion, enabling hybrid walking-driving locomotion strategies. This decomposed motion planner was validated at the DARPA Subterranean Challenge, where the robot rapidly mapped, navigated, and explored underground environments with a higher speed than its traditional legged version. With the lessons learned from this competition, we propose a novel whole-body MPC as a single task formulation that simultaneously optimizes wheel and torso motions. This approach accurately predicts the robot's motion and automatically discovers complex and dynamic movements cumbersome to hand-craft through a decomposed approach. Thanks to the single set of parameters for all behaviors, whole-body optimization makes online gait sequence adaptation possible. Aperiodic gait sequences are automatically found through kinematic leg utilities without the need for predefined contact and lift-off timings. Finding more complex motions over challenging obstacles and at the robot's limits can be achieved through TO methods, optimizing computational-expensive variables like gait timings and considering high-dimensional Centroidal models. By combining offline TO for complex motions and online MPC for continuous optimization along the offline trajectory, we can execute novel maneuvers, including fast motions over challenging obstacles, unique motions through confined spaces, dynamic motions at the robot's limits, and artistic dance moves.