TAMOLS: Terrain-Aware Motion Optimization for Legged Systems

Abstract

Terrain geometry is, in general, non-smooth, nonlinear, non-convex, and, if perceived through a robot-centric visual unit, appears partially occluded and noisy. This work presents the complete control pipeline capable of handling the aforementioned problems in real-time. We formulate a trajectory optimization problem that jointly optimizes over the base pose and footholds, subject to a heightmap. To avoid converging into undesirable local optima, we deploy a graduated optimization technique. We embed a compact, contact-force free stability criterion that is compatible with the non-flat ground formulation. Direct collocation is used as transcription method, resulting in a non-linear optimization problem that can be solved online in less than ten milliseconds. To increase robustness in the presence of external disturbances, we close the tracking loop with a momentum observer. Our experiments demonstrate stair climbing, walking on stepping stones, and over gaps, utilizing various dynamic gaits.

Spatial Group-Wise Enhance: Enhancing Semantic Feature Learning in CNN

“…sparse vegetation occludes the support surface, the vegetation can be filtered out by regarding them outliers. This problem can be tackled by methods using Kalman Filters [4], [12], [13] and other heuristics [7], [14].…”

“…The captured geometrical information can be accumulated in a 2D occupancy map [1], [2], 2.5D elevation map [3], [4], or 3D voxel map [5] representation. These maps provide information about the support surface (the rigid surface that can provide support for the robot during traversing) and obstacles for downstream tasks such as motion planning [6], [7], traversability assessment [8], [9], and trajectory planning [10], [11]. However, these conventional approaches are based on the assumption that the world is rigid, i.e., the robot will step on the terrain or collide with Fig.…”

Editorial: Multimodal behavior from animals to bio-inspired robots

“…The high dynamics control strategy of the bio-inspired robot is impressive and has a history of technology, such as a non-linear program (NLP) (Jenelten et al, 2022 ), Linear Quadratic Regulator (LQR) (Klemm et al, 2020 ), and Model Predictive Control (Di Carlo et al, 2018 ). Besides model-based control, robots driven by neural control architectures, that can respond to internal and external information and implement various behaviors, have superior upper-level capabilities in terms of behavior pattern diversity with smooth transition (Zhu et al, 2022 ) as well as versatile locomotion on complex terrains (Picardi et al, 2020 ; Thor and Manoonpong, 2022 ).…”

Comparison of predictive controllers for locomotion and balance recovery of quadruped robots

“…When the environment is not fully known, it is typically modelled as an elevation map by fusing depth sensor information within proprioceptive information [30], [31]. Recent approaches propose to directly optimise the next contact position, the torso orientation and obstacle avoidance for the foot trajectory based on this input [32], [33], [34]. The approaches share similarities with the framework we propose in terms of the model's proposition.…”