Gasgoo Munich-the DexTeleop TA2 dual-arm robot successfully demonstrated a continuous sequence of tasks—opening a cabinet, grasping objects, and delivering them—during the Intel Technology Innovation and Industry Ecosystem Conference in Suzhou from Sept. 22 to 23. Powered by Intel’s edge server-based embodied intelligence solution, this marked the first time DexTeleop publicly showcased its full-process autonomous task capabilities to ecosystem partners and clients.
Image Credit: DexTeleop
This demonstration was a collaborative effort involving Intel, Xinshu Future, and DexTeleop Intelligence. Intel supplied the on-site computing power via its edge servers, Xinshu Future provided the embodied intelligence "brain" model, and DexTeleop contributed the TA2 robot hardware. During the operation, Xinshu Future’s model handled goal comprehension and decision-making, while the TA2 managed perception, action execution, and hardware-level control and safety. The three parties coordinated to adapt to the specific tasks, successfully bridging the gap from model decision to robotic execution.
Under this embodied intelligence framework, complex perception and reasoning tasks were offloaded to Intel’s edge servers, leaving the TA2 hardware to focus on high-frequency motion control and safety-critical movements. Leveraging Intel’s end-edge heterogeneous computing and software ecosystem, the model’s decisions were translated into the TA2’s physical actions on-site, creating a closed loop from edge computing to robotic execution.
To support model and solution partners in deploying capabilities to the hardware, the DexTeleop TA2 secondary development version offers open interfaces and dedicated technical support. Model developers, research teams, and ecosystem partners can perform secondary development, real-machine debugging, and controlled verification within these open interfaces. Furthermore, they can leverage execution feedback to refine models or optimize upper-layer task logic.
On the data production front, DexTeleop prioritizes multimodal synchronization, result traceability, and a balance between data quality and collection costs. The system ensures strict temporal alignment of images, actions, device status, and task results, while allowing for full traceability of the execution process. This capability helps developers analyze model performance and pinpoint issues, providing a solid data foundation for subsequent training, evaluation, and optimization. Looking ahead, DexTeleop plans to collaborate with model developers, computing partners, and scenario providers to drive further real-machine adaptation and joint scenario validation for a wider range of solutions.
