Embodied AI in Action
Winter 2025/26 to Sommer 2026
In cooperation with Ruhrbot Devils TU Dortmund
Motivation
Autonomous mobile robots must make decisions independently in dynamic environments. Robot soccer is a well-established and practice-oriented benchmark for this: it combines real-time perception, world modeling, motion planning, and decision making under strict timing constraints.
In this setting, robots perceive only a local, limited view of the field through sensors and cameras; fragmented observations must be fused continuously into a consistent internal world model. This world model enables localization of the ball, teammates, and opponents, and it supports strategies and motor actions such as bipedal walking, position changes, and goal shots.
Vision
This project group advances the Nao Devils robot soccer stack by migrating it to a new humanoid platform: instead of the long-used Nao robot, we will work with the Booster Robotics K1, featuring improved mechanics and an integrated NVIDIA Jetson Orin NX GPU platform.
The K1’s on-robot GPU enables modern, compute-intensive AI methods (e.g., vision transformers and learning-based decision making) to run with high inference speed directly on the robot—opening new opportunities for robust, reactive behavior in real time.
Goal
The core objective is a complete human–robot interaction pipeline: a human passes the ball; the robot detects and tracks it, positions itself appropriately, and executes a controlled, targeted shot on goal.
The central challenge is the real-time integration of perception, modeling, motion prediction, and physical execution on the new robot platform.