Overview

Along with 3 other teammates from the State Key Laboratory of Robotics, I participated in the Unmanned Ground Systems Challenge in October 2016 and worked specifically on the environment map building and localization under GPS signal lost situation.

Fig. 1. Our UGV platform

Environment Map

Built an environment map for non-structured fields with the following steps:

Fig. 2. Environment Map BUilding Procedure

The laser data was collected from different lasers: single-line, 32-line, and 64-line. The data was fused after calibration and having manually filled certain points.

Fig. 3. Fuse Laser Data
A road plane was fitted with a RANSAC method.

Fig. 4. Point Clouds Segmentation

The grid map was built with the help of the grid_map package. Finally, the road target was extracted through road skeleton extraction from the image generated from the environment map.

Fig. 5. Road Target Extraction
Thus, we obtained all the perception information needed to drive a car intelligently in a field environment. Obstacle Avoidance Demo

Localization without GPS

Two methods were conducted to meet the requirements.

1. Visual Inertial Odometry

I did this by combining the orb slam with IMU.

Fig. 6. Visual Inertial Odometry

2. Laser Odometry