Mobile robotics
Perception and navigation for mixed-traffic AMRs
Localisation and obstacle handling for autonomous robots sharing aisles with people and forklifts.
- -65%unplanned safety stops
- ±3 cmlocalisation accuracy in aisles
- 10 Hzperception update rate
Stack
- ROS 2
- Ouster 3D LiDAR
- Stereo cameras
- Nav2 planner
- Extended Kalman filter
- Nvidia Jetson
The problem
A Nordic logistics operator ran autonomous mobile robots in a warehouse that also moved manned forklifts and foot traffic. The stock navigation stack stopped often for phantom obstacles and drifted near tall racking, which cut effective throughput and wore down staff trust.
Approach
- 01 Fused 3D LiDAR with wheel odometry and an IMU for localisation that held near featureless racking.
- 02 Added stereo cameras for low obstacles and overhangs the LiDAR missed.
- 03 Classified dynamic agents so the planner slowed for people but flowed around static clutter.
- 04 Tuned the cost map and recovery behaviours against logged shifts rather than a test track.
Outcome
Robots held their lane near tall racking and stopped reacting to dust and reflections. Unnecessary safety stops fell by roughly two thirds and mean distance between manual interventions rose. The operator could add robots to the fleet without hand-tuning each one.
