Mobile Robotics · Sensor Fusion · HRI 2026 · Ongoing

Follow-Me
Assistant Robot

A smart utility robot for domestic and retail environments — following its owner reliably using a fail-over dual-layer tracking system that seamlessly switches between RGB-D visual tracking and UWB radio localization when line of sight is lost.

  • ROS
  • RGB-D
  • UWB
  • NLP
  • Python
  • CAN Bus
  • Scissor Lift
  • Embedded

Project Overview

The Follow-Me Cart is a multi-disciplinary mechatronics project targeting heavy lifting assistance and intuitive user interaction in real-world environments like supermarkets and domestic spaces. The robot is designed to be more than a passive follower — it is an interactive platform with ergonomic, navigational, and communication intelligence.

The key engineering challenge is reliability: ensuring the robot never loses its owner, even in complex environments. This is solved by a dual-layer tracking architecture that uses an RGB-D camera as the primary tracker and UWB radio ranging as a fail-over — seamlessly switching when visual line-of-sight is broken (e.g., behind a store shelf).

Development is currently in the integration and testing phase, with active work on balancing the center of gravity under scissor lift extension and optimising voice-command latency.

Dual-Layer Tracking Architecture

Primary visual tracking via RGB-D camera automatically falls over to UWB trilateration when line of sight is broken — re-acquiring visual lock once the user is back in view.

RGB-D Camera Deep learning person ID 3D obstacle mapping PRIMARY UWB Module Centimetre-level ranging Tag on user · anchors on robot FAILSAFE ↕ failover switch Tracking Arbitrator Line-of-sight monitor LoS active → Visual LoS lost → UWB Re-acquired → Visual Navigation Controller Fused position estimate Path planning Obstacle avoidance Depth map integration DRIVE Motor controllers CAN Bus

Core Features

Four integrated subsystems delivering the full follow-me experience.

Smart Scissor Lift

  • Motorised vertical lift platform
  • Adjustable height based on user preference
  • High-torque linear actuator or lead screw
  • CoG balancing under full load extension

Multimodal Interaction

  • Voice commands via NLP pipeline (follow, stop, go to…)
  • HMI touchscreen for grocery lists and reminders
  • Gesture control for directional nudges
  • Far-field microphone array with noise cancellation

Dual-Layer Tracking

  • RGB-D primary: person ID + 3D position
  • UWB failsafe: centimetre-level ranging
  • Seamless automatic switch on LoS loss
  • Re-acquisition logic to restore visual lock

Intelligent Obstacle Avoidance

  • Real-time depth mapping from RGB-D camera
  • Dynamic obstacle detection and avoidance
  • Handles shoppers, shelves, narrow aisles
  • Integrated with navigation path planner

Current Status & Next Steps

Ongoing — in integration and hardware testing phase.

RGB-D visual person tracking implemented — deep learning-based person identification with 3D position estimation working in controlled environments.

UWB ranging module integrated — centimetre-level tag-to-anchor distance measurement established for failover localization.

Current challenge — balancing the center of gravity when the scissor lift is fully extended under a heavy load. Structural reinforcement under evaluation.

Next — optimising voice-to-command latency in the NLP pipeline and perfecting the UWB re-acquisition logic for real store environments.