Vision-Based Reactive Planning and Control of Quadruped Robots in Unstructured Dynamic Environments

Abstract-- Quadruped robots have received increasing attention for the past few years. For instance, Cheetah can reach a top speed I. INTRODUCTION A controlled backflip is achieved by Panther locomotion capability of quadruped robots like the MIT by representing the rotational dynamics using the rotation Cheetah [1] and ETH Anymal [2]. Still, few prior works focus on solving tasks to find one in daily scenarios, since the dynamic and with complex temporal and logic constraints in unstructured unstructured environment poses significant challenges to and dynamic environments. Hence, this work is motivated to To this end, we consider a robot integrated with vision develop a reactive planning and control strategy for quadruped performing high-level linear temporal logic (LTL) tasks that robots with temporal logic specifications and multi-modal encode position constraints to the target object, as shown in perception to enable mission operation in unstructured Figure 1. To cope with unstructured dynamic environments, we dynamic environments.