AgiBot X2 Humanoid for Elderly Care

Whole-body motion and person-tracking behaviours on the AgiBot X2 humanoid, enabling autonomous following and physical assistance for standing up.

Date
May 2026
Organisation
RoboHack, EPFL · Lausanne
Role
Hackathon team
Technologies
  • ROS 2
  • YOLOv8
  • Stereo depth
  • Whole-body control
  • AgiBot X2
The AgiBot X2 autonomously helping a person stand up at RoboHack, EPFL, followed by the team.

Problem and motivation

Elderly-care assistance: a humanoid that can follow a person and help them stand up. The AgiBot X2’s software stack was early-stage, with limited public documentation.

System

The code is split into two ROS 2 packages:

  • Perception (robot_vision). Real-time stereo person detection with YOLOv8 and per-person stereo depth, publishing the target person’s position.
  • Motion (robot_motion). Head tracking toward the target, a locomotion supervisor that walks toward the person and stops at a set distance, and an arm pose triggered on arrival.

Challenges

Integrating perception with whole-body control on an early-stage stack meant troubleshooting the robot software and iterating on behaviours directly on the physical humanoid.

The AgiBot X2 humanoid on its support frame
The AgiBot X2 during development.
The RoboHack team with the AgiBot X2
The team at RoboHack, EPFL.