Matfyz Humanoid Robot System Excels at WBCD 2026

The prestigious WBCD 2026 international robotics competition took place during the first week of June as part of the IEEE International Conference on Robotics and Automation (ICRA 2026) in Vienna. The team from MFF UK participated in a challenge focused on shelf picking – the autonomous or assisted search and retrieval of goods from shelves, which is currently one of the most technologically sought-after disciplines for logistics and automated warehouses.

In this discipline, the robot is controlled remotely by a human operator wearing motion-capture sensors, who monitors the situation through a VR headset. This hybrid concept, where human intuition and fine motor skills help overcome the limitations of purely autonomous decision-making in unfamiliar environments, earned the Matfyz team 2nd place in strong international competition.

The entire system was designed and programmed by robotics student Xuzhe Dang, from motion capture via reinforcement learning to the full-body control of the robot. Directly on-site in Vienna, the machine was controlled remotely by Kryštof Pšenička. Other students also participated in the project, testing partial solutions such as stereo cameras, motion capture sensors, VR headsets, or two precise robotic hands.

„We were even reaching for the gold medal. After the first run, we were leading with 96 points and thought that only the best result was scored, not the sum of both runs. In the second run, we therefore chose a far more risky strategy. This eventually didn't pay off as the robot fell down, but from a scientific perspective, we are thrilled: our 96 points from the first run remained unbeaten, and our system proved to be the most versatile and robust of all,“ commented Dr. Stefan Edelkamp from the Department of Theoretical Computer Science and Mathematical Logic at MFF UK, who led the team.

Several other team members based in Prague also contributed to the work, including David Obdržálek, Ibrahim Qatato, Amal Aaramam Sudhikumar, and Nathan Vayva, an intern in the Robot Lab.

The teleoperation system consists of several components: a whole-body control policy that controls the humanoid robot, excluding the robotic hands and camera gimbal; a hand control algorithm based on motion mapping (retargeting); a camera system for transmitting video streams and controlling the gimbal, which ensures gaze direction tracking in virtual reality; and a Unity-based VR application that tracks the operator’s motions and sends commands to the robot.

The system is modular, allowing each component to operate independently and adapt to different methods. „The whole-body control policy is either trained in our own framework, or we use an open-source policy, such as SONIC. The algorithm for the robotic hands currently mainly supports different types of five-finger hands. As a VR platform, either Quest 3 or PICO can be used,“ specified Xuzhe Dang. „Thanks to the WBCD competition, we immediately identified new research directions in our current remote control system. One major issue is that the whole-body control policy is not sufficiently sensitive to less pronounced movements from the operator. We consider the integration of visual data into our control strategy to be an important step toward our ultimate goal of building a fully autonomous humanoid robot,“ the young researcher adds.

MFF UK Robotic Lab

The MFF UK Robotic Lab, which operates under the banner of the Department of Theoretical Computer Science and Mathematical Logic (KTIML), is headed by Prof. Roman Barták. The facility itself and the acquisition of top-tier hardware are developing primarily under the leadership of Dr. Stefan Edelkamp and Dr. David Obdržálek, thanks in part to the support of the Donatio Universitatis Carolinae grant. The laboratory focuses on software development and control systems. Scientists here address areas such as automatic behavior planning, energy-efficient movement in space, and the mutual coordination of multiple robots when performing a single task.