Robotics & perception

A way forward.

Autonomous mobile robot navigation

Computer vision and LiDAR help mobile robots navigate changing warehouse environments.

YF / 02 Robotics & perception Autonomous mobile robot navigation — conceptual system illustration
System concept Illustration / 02
Industry
Logistics & fulfilment
Our role
Perception & autonomous navigation
Core tools
Jetson AGX Orin · LiDAR · ROS 2
01 / The challenge

A practical problem.

Fixed-path robots stopped whenever an unexpected object blocked their route. Changing warehouse layouts and shared aisles called for a system that could detect obstacles and plan around them.

Engagement
8 months · Jun 2023–Jan 2024
Project setting
Three distribution centres · Yorkshire
02 / Our approach

Built around the constraints.

  1. Combine the sensors.

    Camera, LiDAR, and motion data provide a shared view of the robot’s surroundings.

  2. Understand the route.

    An optimised detection model runs on Jetson, while SLAM keeps the robot localised.

  3. Adapt as things change.

    Dynamic planning updates routes around obstacles and connects to warehouse task dispatch.

03 / The system

How it comes together.

Components → processing → output

On-robot perception and planning connect sensor data to a continuously updated route through the warehouse.

  1. Sense

    Stereo cameras · LiDAR · IMU

    Camera frames, laser scans, and motion readings describe nearby obstacles and the robot’s movement.

    OutputVisual, depth & motion data

  2. Localise

    Kalman filtering · SLAM

    Fused sensor readings estimate the robot’s pose. Mapping and obstacle tracking maintain its view of the surroundings.

    OutputRobot pose & obstacle map

  3. Plan

    A* · Dynamic Window Approach

    A global planner selects the route; the local planner updates motion around obstacles at 10 Hz.

    OutputUpdated route & velocity targets

  4. Move

    Motion controller · Stop system

    The controller follows the planned trajectory with smooth acceleration. Stop controls and manual override remain available.

    OutputRobot movement & state feedback

04 / The outcome

Results from this project.

Obstacle detection
99.7%Measured on dry surfaces
Detection latency
15 msOn NVIDIA Jetson AGX Orin
Localisation accuracy
2 cmFor precise positioning and docking

Before

Fixed-path robots halted at unexpected obstacles and needed manual clearance.

After

Sensor fusion and dynamic planning let robots adapt to obstacles and changing layouts.

In context. Wet floors can reduce LiDAR performance. Continuous operation is limited to around 6.5 hours per charge, with charging planned into the workflow.

Your next project

Working on something similar?

Tell us about your system and where you need support.

Discuss your project