Gasgoo Munich- At the 2026 World Artificial Intelligence Conference (WAIC 2026), PNDbotics showcased its Adam full-size humanoid and Adam-U upper-body robot. The booth featured demonstrations ranging from motion capture and teleoperation to stair climbing and dual-model VLA picking — illustrating a complete technology loop from data collection and model validation to physical execution.
A full-stack developer and manufacturer in the humanoid space, PNDbotics has mastered core technologies including gait planning, whole-body coordination, and deep reinforcement learning (DRL) motion control. The company has achieved full-stack in-house development of both core components and complete units, launching the full-size Adam Pro and the upper-body Adam-U. With dual smart manufacturing hubs in Ningbo and Tianjin, the company is now accelerating mass production and deliveries.
Image Credit: PNDbotics
At the PNDbotics booth, the Adam full-size humanoid took center stage. Standing 167 cm tall with 43 degrees of freedom, the robot demonstrated its stair-climbing capabilities — achieving stable walking and dynamic balance on a simulated staircase. The display verified its motion control and robustness in unstructured environments.
Notably, this marked the first public demonstration of Adam’s stair-climbing ability in an exhibition setting.
Simultaneously, Adam showcased motion capture and teleoperation capabilities. Using a motion capture system, an operator's movements were mapped onto the robot in real time, enabling high-precision, low-latency trajectory tracking.
Joining Adam was the Adam-U upper-body robot. Two units were displayed side-by-side, each running a different embodied intelligence model to visually demonstrate the platform's openness and adaptability. Both Adam-U units performed VLA picking tasks simultaneously — driven by natural language commands, each model independently handled target recognition, task planning, and execution.
Adam-U also served as a platform for motion capture data collection. By efficiently gathering real-world data via the motion capture system, it provided high-quality "fuel" for training reinforcement learning and imitation learning models — directly addressing the industry's bottleneck of data scarcity in embodied intelligence.
Beyond the robots themselves, PNDbotics displayed core components including motors, drivers, and reducers. This highlighted its full-stack in-house capabilities, spanning from underlying hardware to upper-layer algorithms.
Notably, the proprietary PND-Network real-time communication system offers ultra-high bandwidth support for over 100 to 200 degrees-of-freedom nodes. This provides the technical foundation for multi-robot collaboration and large-scale deployment.
