Robust Rough-Terrain Locomotion with a Quadrupedal Robot

Abstract

Robots working in natural, urban, and industrial settings need to be able to navigate challenging environments. In this paper, we present a motion planner for the perceptive rough-terrain locomotion with quadrupedal robots. The planner finds safe footholds along with collision-free swing-leg motions by leveraging an acquired terrain map. To this end, we present a novel pose optimization approach that enables the robot to climb over significant obstacles. We experimentally validate our approach with the quadrupedal robot ANYmal by autonomously traversing obstacles such steps, inclines, and stairs. The locomotion planner re-plans the motion at every step to cope with disturbances and dynamic environments. The robot has no prior knowledge of the scene, and all mapping, state estimation, control, and planning is performed in real-time onboard the robot.

Walking Posture Adaptation for Legged Robot Navigation in Confined Spaces

“…Unlike [18], as our abstraction changes, the complexity of the planning problem remains the same which allows for fast computation in confined spaces. We do not plan the foot tip locations, however our method could later be used with a leg swing planner such as in [11] to avoid any collision of the robot's legs with the terrain.…”

“…Weaver, for example, uses proprioceptive terrain characterisation and an admittance controller [10] for uneven terrain. ANYmal uses robot-centric elevation mapping to select individual footholds that are both on suitable terrain and satisfy kinematic constraints [11]. However, neither of these methods adapt the robot's posture for walking through confined environments.…”

DeepGait: Planning and Control of Quadrupedal Gaits Using Deep Reinforcement Learning

“…To this end, we employ Deep Reinforcement Learning (DRL) techniques to train an agent comprising a two-layer hierarchy of Neural-Network (NN) policies, which partitions locomotion into separate components responsible for foothold planning and tracking control respectively. Such problems have predominantly been addressed using state-of-the-art Trajectory Optimization (TO) techniques [2], [3] as well as other model-based approaches [4], [5]. However, as they require several modeling assumptions and approximations, they consistently present trade-offs between computational efficiency and scalability.…”

Robust Legged Robot State Estimation Using Factor Graph Optimization

“…For statically stable walking, leg odometry drift is low enough that terrain mapping can be used for continuous footstep planning and execution [4]. However for dynamic locomotion, position drift is much higher which makes such mapping ineffective, as illustrated in Fig.…”