Obstacle-Avoidant Leader Following with a Quadruped Robot

TL;DR

This work tackles autonomous leader-following for a quadruped robot in dynamic environments to reduce operator effort. It introduces a multimodal pipeline that fuses LiDAR, RGB-D cameras, and a novel Angle of Arrival AoA sensor for robust leader detection and tracking via an Extended Kalman Filter, supplemented by a waverider-based local planner for dynamic obstacle avoidance. Key contributions include the AoA beacon, tight sensor fusion, and real-time SE(2)-level navigation that handles crowds and occlusions, demonstrated on the ANYmal platform. The approach enables safer, hands-free operation in industrial and assistive scenarios, with potential to extend personal mobility and site inspection tasks.

Abstract

Personal mobile robotic assistants are expected to find wide applications in industry and healthcare. For example, people with limited mobility can benefit from robots helping with daily tasks, or construction workers can have robots perform precision monitoring tasks on-site. However, manually steering a robot while in motion requires significant concentration from the operator, especially in tight or crowded spaces. This reduces walking speed, and the constant need for vigilance increases fatigue and, thus, the risk of accidents. This work presents a virtual leash with which a robot can naturally follow an operator. We use a sensor fusion based on a custom-built RF transponder, RGB cameras, and a LiDAR. In addition, we customize a local avoidance planner for legged platforms, which enables us to navigate dynamic and narrow environments. We successfully validate on the ANYmal platform the robustness and performance of our entire pipeline in real-world experiments.

Figures (5)

• Figure 1: We deploy our full pipeline on the ANYmal robot and show resilient leader following in crowded and complex environments. The robot identifies the leader through a beacon they carry, emphasized with blue, the signal of which is picked up by a custom antenna array on the robot.
• Figure 2: The presented pipeline consists of three main elements: the sensor fusion (A), leader tracking (B), and leader following (C). We combine measurements from the onboard cameras and LiDAR unit of the robot with an additional custom AoA sensor unit. From these measurements, we segment our leader out of the scene, which may involve other people. Using an EKF, we track the leader's motion, which allows us to keep track of them, even when occluded. By adding a new waverider policy for dynamic obstacles, the robot can follow the leader through crowded spaces, avoiding collision with both the environment and (potentially moving) other people.
• Figure 3: Our novel AoA Sensor composed of a SDR, SM and an antenna array on top.
• Figure 4: We modify waverider to track our leader while autonomously avoiding static obstacles, such as benches and door frames. We create an occupancy map with wavemap, actively filtering out human points to only include the static scene. The map resulting from a tracking sequence can be seen in (A). Said sequence is visualized in (B), with only the final leader frame visualized to reduce visual clutter. In (C), we show the path the robot takes while avoiding collisions. The leader trajectory (as recorded by the robot) is visualized via the purple line and the robot trajectory via the sequence of arrow clusters. Each cluster represents a pose and the forces acting on the robot during that single time step, the green symbolizing the attractive force of the goal and the red the repelling forces of the obstacles within range. The image is split into three frames to highlight the multi-resolution nature of waverider, with the green blocks representing the obstacles. Each third of the image represents one timestep, and the black cross shows the exact robot position for that timestep.
• Figure 5: In a situation with two people present, the pipeline correctly segments out both (green bounding boxes), and identifies its leader from the two. The leader point cloud is additionally highlighted in green. When the leader starts to move, as shown on the right, the robot follows along, diverging to the left to avoid collision with the other human.