Work / Mobile robotics, HRI Hardware
Multi-floor receptionist robot
A semi-humanoid that guides visitors across a whole building
Most indoor mobile robots are limited to a single 2D map. This project extended standard 2D SLAM and navigation into a full multi-floor framework, allowing a receptionist robot to guide visitors across an entire building. The robot is built on a custom sheet-metal base carrying an RPLIDAR A2 and an Intel RealSense D435, with an NVIDIA Jetson AGX Xavier as onboard compute.
Challenge
Elevators break the assumptions of 2D navigation: the robot must call, wait, enter, ride, recognise the floor and switch maps, reliably and around people.
Approach
Floor transitions are sequenced by behaviour trees (BehaviorTree.CPP) covering the full elevator routine. The whole routine was first developed in Gazebo with custom elevator plugins written for the project. A ReSpeaker microphone array provides voice interaction and directional listening; a human-aware navigation layer, strengthened with a replicated human goal-prediction method, lets the robot move around people without intruding. MoveIt plans the upper-body arm.
Result
A hardware prototype tested in real buildings, with the software split into modular ROS 2 packages for navigation, perception, speech, behaviour and HMI.
In brief
- Multi-floor navigation framework built on 2D SLAM and Nav2
- Custom Gazebo plugins for elevator simulation
- Behaviour-tree automation of the complete elevator sequence
- Human goal prediction to improve crowd navigation
- Voice interaction through a ReSpeaker microphone array
- Trade-offs worked through: cloud LLM vs local model, QoS tuning for critical sensors, behaviour trees over state machines, LiDAR SLAM over visual SLAM