Exteroceptive perception for AlienGo and corresponding downstream applications

Honghao Liao

VsisLab, Shandong University, China

202214818@mail.sdu.edu.cn

Project Duration: 2022.10 - 2023.06

Project Introduction

In this project, we implement an exteroceptive sensory data acquisition framework for the quadruped robot Unitree AlienGo. The framework reads sensor data from the robot’s built-in depth camera D435, visual odometer T265, and external lidar VLP-16. Furthermore, the sensor data can be packaged and transferred between devices in the form of ROS topics. With the framework, we subsequently implemented three downstream tasks: SLAM and autonomous navigation for quadruped robot, elevation map-based navigation system for quadruped robot, and an experimental project footage planning for quadruped robot.

Demonstrations for Downstream Tasks

In this section, we will give a brief overview of the downstream tasks showed in this page, in the form of demonstration videos and diagrams.

SLAM and Autonomous Navigation

For SLAM task, we use the Gmapping and the LIO-SAM to achieve the construction of 2D planar maps and 3D point cloud maps of the surroundings by the quadruped robot.

alt text
Indoor large-scene 3D point cloud map construction (total area of about 1000 square metres). The blue line in the figure is the movement route of the quadruped robot during map building.
alt text
3D point cloud map construction for indoor scene with long corridor loop (total length of corridor loop is approximately 80m).
alt text
Indoor large scene 2D map construction. The green line in the figure is the movement route of the quadruped robot during map building.

For the autonomous navigation task, we implemented a navigation system for quadruped robots based on ROS Navigation Stack.

demonstration video for autonomous navigation system.

Elevation Map-based Navigation System

In this task, we used elevation mapping cupy as an elevation map construction tool. Further, we implemented an autonomous navigation system based on elevation maps, which contains global path planning and local path planning. This system can help the quadruped robot bypass or directly cross different obstacles according to its own locomotion ability.

alt text
top-down view of elevation map example.
alt text
lateral view of elevation map example.
demonstration video for elevation map-based navigation system.

Footage Planning

The task is done based on quad-sdk and focuses mainly on the implementation of footage planning for quadruped robots Unitree AlienGo using surrounding elevation information. Note that some of the following demos incorporate the global path planner gbpl.

Climb up a 20cm high stair.
Climb slope.
Walk cross a 20cm wide ditch.
Jump up a 20cm high stair(with global planner).
Leap over a 20cm wide ditch(with global planner).