RoboNaldo: Accurate, Stable and Powerful Humanoid Soccer Shooting via Motion-Guided Curriculum Reinforcement Learning

2026-06-12 · Source: cs.AI updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, long

Summary

RoboNaldo is a three-stage motion-guided curriculum reinforcement learning framework designed for high-impulse humanoid soccer shooting, enabling accurate, stable, and powerful kicks. It utilizes a single human-kick reference as a scaffold, gradually optimizing for shooting performance. The curriculum first establishes a stable whole-body kicking prior, then adapts the kick for stationary-ball free-kick scenarios with random ball positions, and finally extends it to moving-ball shooting through a locomotion-command and kick-trigger interface. In simulation, RoboNaldo achieved a 48.6% lower free-kick shot error and 2.96× higher shoot velocity compared to prior baselines, with Stage 2 free-kicks showing 0.899 m average error from 5 m and 14.79 m/s ball speed. On a real Unitree G1 robot with onboard perception, it demonstrated 0.73 m and 0.86 m average target shooting error from 3 m in free-kick and moving-ball cases, respectively, with ball speeds up to 13.10 m/s.

Key takeaway

For Robotics Engineers developing athletic humanoid interactions, particularly high-impulse tasks like kicking, you should adopt a staged curriculum reinforcement learning approach. This method, which progressively builds from stable motion priors to task-specific adaptations and precise timing, significantly improves both stability and accuracy. Implement proximity-based tracking relaxation and robust perception to achieve real-world performance, enabling your robots to execute complex, dynamic maneuvers effectively.

Key insights

A staged curriculum combining motion tracking and task rewards effectively teaches complex, high-impulse humanoid interactions like soccer shooting.

Principles

• Combine motion priors for stability with task rewards for adaptability.
• Decompose complex tasks into sequential, learnable stages.
• Relax motion tracking constraints only where task objectives conflict.

Method

RoboNaldo employs a three-stage curriculum: first, motion tracking learns a stable kick prior; second, shooting rewards adapt the kick for accurate stationary-ball free-kicks; third, a locomotion-command/kick-trigger interface enables moving-ball shooting.

In practice

• Implement proximity-based tracking relaxation for precise contact control.
• Utilize retro-reflective properties for robust ball detection via LiDAR and IR cameras.
• Apply extensive domain randomization for effective sim-to-real transfer.

Topics

• Humanoid Robotics
• Reinforcement Learning
• Curriculum Learning
• Soccer Shooting
• Whole-Body Control
• Sim-to-Real Transfer

Best for: Research Scientist, AI Scientist, Robotics Engineer, Machine Learning Engineer

Related on AIssential

• Mind Your Steps: A General Learning Framework for Accurate Humanoid Foothold Tracking — The framework enables terrain-agnostic 3D humanoid foothold tracking by dynamically sampling goals and using a robust target representation for real-world transfer.
• A history of RoboCup with Manuela Veloso — RoboCup's evolution reflects AI and robotics research trends, now focusing internationally on humanoid soccer for visibility and accessibility.
• Video Friday: Digit Learns to Dance—Virtually Overnight — Advanced robotics are achieving new performance thresholds in control, learning, and human interaction.
• Training a Robotic Arm Using MuJoCo and JAX on AMD Hardware with ROCm™ — Reinforcement learning with MuJoCo and JAX on AMD ROCm enables robust robotic arm training for complex manipulation tasks.
• AI-powered robot beats elite table tennis players — Sony AI's Ace robot demonstrates advanced real-time perception and control in complex human-robot interaction.
• Why Do Humanoid Robots Still Struggle With the Small Stuff? — Despite AI advancements, humanoid robots still struggle with fundamental physics like force control for reliable real-world interaction.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.AI updates on arXiv.org.