Gem e2 demo

LiDAR-based obstacle detection with grid density clustering, geometric waypoint planning, and pure pursuit GPS tracking on a Polaris GEM autonomous vehicle.
Code Video

Overview

This project implements an autonomous navigation pipeline on a Polaris GEM e2 vehicle using a Velodyne LiDAR sensor. The vehicle detects two physical boxes in its environment, plans a path through the gap between them, and drives to the midpoint using GPS-guided pure pursuit control — all within a ROS framework.

System Architecture

Velodyne LiDAR
PointCloud2 stream
→
Box Locator
Grid density clustering
→
Waypoint Planner
Midpoint + GPS transform
→
Pure Pursuit
PACMod vehicle control
ROS pipeline on Polaris GEM e2 — Novatel GPS + Velodyne LiDAR

LiDAR Box Detection

The Velodyne point cloud is filtered to retain only points in the forward detection zone — within 2m laterally and at a minimum distance from the vehicle. The surviving points are projected onto a 12×12 occupancy grid in the XY plane, and the two highest-density cells are identified as box centers.

$$ \text{density}(i,j) = \sum_{\mathbf{p} \in \text{cell}(i,j)} \mathbf{1}[-1.4 \le p_z \le 0] $$

The two peak-density blocks are returned as the detected obstacle positions.

Waypoint Planning

The planner targets the midpoint between the two boxes as the navigation goal, then linearly interpolates 15 waypoints from the vehicle’s current position to that target:

$$ \mathbf{w}_i = \mathbf{p}_{\text{curr}} + \frac{i}{N-1}(\mathbf{p}_{\text{dest}} - \mathbf{p}_{\text{curr}}), \quad i = 0, \dots, N-1 $$

Waypoints are then rotated and translated from LiDAR-local frame into GPS world coordinates using the vehicle’s heading from the Novatel INS.

Results

Component Method
Obstacle DetectionVelodyne LiDAR + grid density clustering
LocalizationNovatel GPS/INS (NovatelPosition + Inspva)
Path PlanningLinear interpolation to gap midpoint
ControlPure pursuit via PACMod interface
PlatformPolaris GEM e2, ROS Noetic

Key Observations