D-Robotics and Infinigence AI Team Up on an Embodied Intelligence Development Platform

According to leaderobot.com, D-Robotics and Infinigence AI announced on November 13, 2026, a collaboration linking open-source platforms for embodied intelligence with D-Robotics’ computing platform and software. The platforms are RLark, RLinf, RPent and APXInf; the D-Robotics products are its Sunrise series chips and RoboGo software platform. The report describes a setup supporting the process from training to agent execution and deployment on devices.

The reported collaboration lays out a path for managing four Sunrise models with RLark, and for sending tasks from RPent to physical boards and collecting trajectory data. This gives developers a concrete option for connecting training environments with on-device computing resources, but the supported models and performance figures alone do not establish reproducibility in real-world operations or the scope of potential applications.

The focus of the partnership is not the announcement of a new robot, but linking computing resources and operational software across the distinct stages of embodied intelligence development: environment management, training, agent execution and device deployment. In the setup described in the report, RLark handles environment and equipment management, RLinf handles training, RPent handles agent execution, APXInf handles deployment, and the Sunrise series provides on-device computing. The effort is therefore not simply about expanding chip and software compatibility; it aims to put the steps from cloud-based development to execution on physical boards and trajectory-data collection on the same set of platforms. D-Robotics was also covered in D-Robotics announces RawVLA, leaderobot.com reports, published on November 4, 2026. The article summary says it also introduced chips, development kits and software platforms. The current report covers how those platforms are being connected to different stages of model development. Meanwhile, AgiBot, Tsinghua University and others announce CEL, test swarm learning with 96 robots is a separate initiative, published on November 2, 2026, that also involved Infinigence AI. The results of CEL’s physical-robot tests and the implementation status of the platform collaboration reported here should be evaluated as separate developments, even though the same company appears in both. The broader industry implication is that a development platform’s usefulness depends not only on hardware performance, but also on whether tools for different stages can connect, call on physical resources remotely and return execution results. In particular, support for four Sunrise models in RLark and RPent’s connection to on-device sandboxes were reported as concrete examples linking equipment management with physical execution. However, model compatibility alone does not show portability to different robots or configurations, or stability in sustained operation. The reported speedup of up to 2.43x also cannot be compared with results in other environments without knowing the configuration and conditions under which it was measured. Key points to verify include the range of equipment and tasks used, performance-measurement conditions, how collected trajectory data is handled, and on-device safety.