On Terrain-Aware Locomotion for Legged Robots

Legged robots are advancing towards being fully autonomous as can be seen by the recent developments in academia and industry. To accomplish breakthroughs in dynamic whole-body locomotion, and to be robust while traversing unexplored complex environments, legged robots have to be terrain aware. Terrain-Aware Locomotion (TAL) implies that the robot can perceive the terrain with its sensors, and can take decisions based on this information. The decisions can either be in planning, control, or in state estimation, and the terrain may vary in geometry or in its physical properties. TAL can be categorized into Proprioceptive Terrain-Aware Locomotion (PTAL), which relies on the internal robot measurements to negotiate the terrain, and Exteroceptive Terrain-Aware Locomotion (ETAL) that relies on the robot’s vision to perceive the terrain. This thesis presents TAL strategies both from a proprioceptive and an exteroceptive perspective. The strategies are implemented at the level of locomotion planning, control, and state estimation, and are using optimization and learning techniques. The first part of this thesis focuses on PTAL strategies that help the robot adapt to the terrain geometry and properties. At the Whole-Body Control (WBC) level, achieving dynamic TAL requires reasoning about the robot dynamics, actuation and kinematic limits as well as the terrain interaction. For that, we introduce a Passive Whole-Body Control (pWBC) framework that allows the robot to stabilize and walk over challenging terrain while taking into account the terrain geometry (inclination) and friction properties. The pWBC relies on rigid contact assumptions which makes it suitable only for stiff terrain. As a consequence, we introduce Soft Terrain Adaptation aNd Compliance Estimation (STANCE) which is a soft terrain adaptation algorithm that generalizes beyond rigid terrain. STANCE consists of a Compliant Contact Consistent Whole-Body Control (c3WBC) that adapts the locomotion strategies based on the terrain impedance, and an online Terrain Compliance Estimator (TCE) that senses and learns the terrain impedance properties to provide it to the c 3WBC. Additionally, we demonstrate the effects of terrains with different impedances on state estimation for legged robots. The second part of the thesis focuses on ETAL strategies that makes the robot aware of the terrain geometry using visual (exteroceptive) information. To do so, we present Vision-Based Terrain-Aware Locomotion (ViTAL) which is a locomotion planning strategy. ViTAL consists of a Vision-Based Pose Adaptation (VPA) algorithm to plan the robot’s body pose, and a Vision-Based Foothold Adaptation (VFA) algorithm to select the robot’s footholds. The VFA is an extension to the state of the art in foothold selection planning strategies. Most importantly, the VPA algorithm introduces a different paradigm for vision-based pose adaptation. ViTAL relies on a set of robot skills that characterizes the capabilities of the robot and its legs. These skills are then learned via self-supervised learning using Convolutional Neural Networks (CNNs). The skills include (but are not limited to) the robot’s ability to assess the terrain’s geometry, avoid leg collisions, and to avoid reaching kinematic limits. As a result, we contribute with an online vision-based locomotion planning strategy that selects the footholds based on the robot capabilities, and the robot pose that maximizes the chances of the robot succeeding in reaching these footholds. Our strategies are extensively validated on the quadruped robots HyQ and HyQReal in simulation and experiment. We show that with the help of these strategies, we can push dynamic legged robots one step closer towards being fully autonomous and terrain aware.