ViNL: Visual Navigation and Locomotion Over Obstacles

ARMP: Autoregressive Motion Planning for Quadruped Locomotion and Navigation in Complex Indoor Environments

“…Although promising results have been presented for wheel-based robots [14]- [17], the navigation of legged robots is still at a premature stage without considering joint-level planning. Therefore, legged robot navigation is usually tackled using a hierarchical structure where the high-level controller generates a command variable for a low-level controller [18], [19] who runs at a higher frequency. The low-level controller can be either expert designed [18]- [20] or learned using reinforcement learning [21], [22].…”

“…Therefore, legged robot navigation is usually tackled using a hierarchical structure where the high-level controller generates a command variable for a low-level controller [18], [19] who runs at a higher frequency. The low-level controller can be either expert designed [18]- [20] or learned using reinforcement learning [21], [22]. However, most of the previous works only considers environments with flat surfaces, which do not fully exploit the potential of the legged robot.…”

“…However, most of the previous works only considers environments with flat surfaces, which do not fully exploit the potential of the legged robot. Although a recent work [18] addresses the legged robot navigation with small obstacles scattered on the surface, the obstacles are only treated as disturbances. In this work, our goal is to generate a feasible motion plan for a legged robot in realistic indoor environments that include stairs and obstacles.…”

“…This makes ARMP an effective low-level motion generator for indoor navigation with complex terrain settings. For all the experiments, highlevel trajectories are generated manually or using the offthe-shelf motion planner such as RRT* because learning a navigation policy is not the main focus of the paper, although it can be approached by other visual navigation techniques [18]. In this section, we investigate ARMP in the following three scenarios.…”