Hierarchical Experience-informed Navigation for Multi-modal Quadrupedal Rebar Grid Traversal

TL;DR

This work addresses kinodynamically feasible navigation for a quadruped across constrained rebar grids, a problem that combines discrete footstep decisions with continuous dynamics. It introduces a hierarchical planner that couples a mode-transition graph with a kinodynamic trajectory optimizer, augmented by an experience-based edge weighting and a guiding torso path to improve global planning. Key contributions include adapting ALEF's mode-family foliations to quadruped contact planning, integrating offline experience into edge weights via a tanh-based mapping and RBF updates, and validating the approach in both simulation and hardware. The results demonstrate improved planning efficiency, reduced offline trial requirements, and robust navigation around obstacles, with clear potential for online deployment and extension to additional gaits and perception-enabled environments.

Abstract

This study focuses on a layered, experience-based, multi-modal contact planning framework for agile quadrupedal locomotion over a constrained rebar environment. To this end, our hierarchical planner incorporates locomotion-specific modules into the high-level contact sequence planner and solves kinodynamically-aware trajectory optimization as the low-level motion planner. Through quantitative analysis of the experience accumulation process and experimental validation of the kinodynamic feasibility of the generated locomotion trajectories, we demonstrate that the experience planning heuristic offers an effective way of providing candidate footholds for a legged contact planner. Additionally, we introduce a guiding torso path heuristic at the global planning level to enhance the navigation success rate in the presence of environmental obstacles. Our results indicate that the torso-path guided experience accumulation requires significantly fewer offline trials to successfully reach the goal compared to regular experience accumulation. Finally, our planning framework is validated in both dynamics simulations and real hardware implementations on a quadrupedal robot provided by Skymul Inc.

Figures (8)

• Figure 1: Illustration of a quadrupedal robot performing rebar grid traversal in a simulated construction environment clearpath_robotics_clearpath_nodate and an indoor test-bed.
• Figure 2: Kinematic reachability areas along with projected configuration samples. Some samples that meet the distance threshold are set as outliers to avoid reachable areas that yield unstable contacts or self-collisions.
• Figure 3: Overall diagram of planning framework. A graph search (Section \ref{['sec:graph_search']} is performed over mode transitions using the estimated weight distributions from experience (Section \ref{['sec:experience']}) which are colored by their corresponding swing foot. Then, the suggested contact sequence is run through trajectory optimization (Section \ref{['sec:traj_opt']}) to determine the costs $\mathcal{J}_{i, i+1}$ of the transitions.
• Figure 4: Rebar grid layouts used in the case studies in Sections \ref{['sec:cs1_exp']} - \ref{['sec:cs3_construction']}.
• Figure 5: Results for Section \ref{['sec:cs1_exp']}: Case Study 1 on the grid in Figure \ref{['fig:exp2_setups']}(a). A gradual increase in progress through the proposed contact sequences can be seen in left figure as the graph search comes to avoid the more difficult contact transitions that lead to collisions. Once successful contact sequences are identified through experience, the graph search time drops significantly.