Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Rhoban Football Club: RoboCup Humanoid Kid-Size 2023 Champion Team Paper

Julien Allali, Adrien Boussicault, Cyprien Brocaire, Céline Dobigeon, Marc Duclusaud, Clément Gaspard, Hugo Gimbert, Loïc Gondry, Olivier Ly, Grégoire Passault, Antoine Pirrone

TL;DR

This paper documents Rhoban Sigmaban’s 2023 RoboCup Kid-Size champion framework, detailing a revamped walk engine, enhanced perception, and strategy-monitoring pipelines. The walk system combines RL-based footsteps planning with a $7$-step CoM/IK cascade and a QP-based whole-body IK, enabling stable, fast locomotion while accommodating kick and stand-up actions via keyframes. Perception is upgraded with a higher FOV lens and end-to-end YOLOv8 detection, trained on a curated, augmented dataset and fused with odometry and foot-pressure cues for reliable localization. The strategy and monitoring stack coordinates kick and placement decisions through a Captain-driven communication protocol and a Python-based debugging/replay suite, contributing to the team’s competitive performance and openness to future open-source release.

Abstract

In 2023, Rhoban Football Club reached the first place of the KidSize soccer competition for the fifth time, and received the best humanoid award. This paper presents and reviews important points in robots architecture and workflow, with hindsights from the competition.

Rhoban Football Club: RoboCup Humanoid Kid-Size 2023 Champion Team Paper

TL;DR

This paper documents Rhoban Sigmaban’s 2023 RoboCup Kid-Size champion framework, detailing a revamped walk engine, enhanced perception, and strategy-monitoring pipelines. The walk system combines RL-based footsteps planning with a -step CoM/IK cascade and a QP-based whole-body IK, enabling stable, fast locomotion while accommodating kick and stand-up actions via keyframes. Perception is upgraded with a higher FOV lens and end-to-end YOLOv8 detection, trained on a curated, augmented dataset and fused with odometry and foot-pressure cues for reliable localization. The strategy and monitoring stack coordinates kick and placement decisions through a Captain-driven communication protocol and a Python-based debugging/replay suite, contributing to the team’s competitive performance and openness to future open-source release.

Abstract

In 2023, Rhoban Football Club reached the first place of the KidSize soccer competition for the fifth time, and received the best humanoid award. This paper presents and reviews important points in robots architecture and workflow, with hindsights from the competition.
Paper Structure (27 sections, 3 equations, 8 figures)

This paper contains 27 sections, 3 equations, 8 figures.

Table of Contents

  1. Robots architecture
  2. Overview
  3. Software
  4. Recent mechanical upgrades
  5. Bumpers
  6. Hip yaw bearings
  7. Vision, perception and localisation
  8. Lenses and Field of view
  9. Camera calibration
  10. Image processing
  11. Training:
  12. On-robot inference:
  13. Dataset
  14. Images selection and data augmentation:
  15. Detected features:
  16. ...and 12 more sections

Figures (8)

  • Figure 1: Inference on a frame with our trained neural network
  • Figure 2: Normalized confusion matrix of our neural network (confidence threshold=0.25)
  • Figure 3: Overview of the stages of the walk engine
  • Figure 4: Cartesian trajectory of a swinging foot of $5cm$ along $x$ axis and $5cm$ along $y$ axis
  • Figure 5: Planned velocities during the walk for the knee and ankle actuators. The red area represent the operational range of the actuators given by the ROBOTIS for MX-106 Robotis-MX106.
  • ...and 3 more figures