Unitree Robotics Dir: Is Embodied AI Better with More Autonomy? 5-Layer On-Demand Adaptation

Описание: At the GOSIM HANGZHOU 2025 venue, the program Open AGI Forum turns its lens to Unitree Robotics, one of "Hangzhou's Six Dragons". It delves into how the company integrates large models into quadruped/humanoid robots, enables robot dogs to make deliveries, and deploys humanoid robots for screw-tightening tasks in factories. It also explores their reflections on "the final mile to AGI".

This episode features ShengYue Ma, Robot Engineer/Technical Director,Unitree Robotics, in a in-depth conversation with Echo Tang, Executive Editor-in-Chief of CSDN & New Programmer. The dialogue uncovers Unitree's practices and insights in robot deployment, AGI breakthroughs, and open-source ecosystems, packed with valuable information that is worth saving for careful reading!

Video Highlights Quick Glance

·From the Department of Intelligence at Sun Yat-sen University to Unitree's CTO: What career path led to "closing the algorithm loop on hardware"?

·Analyzing Unitree's technological competitiveness: Full independent R&D of software and hardware achieves "early problem detection and late testing closure".

·How to leverage the dual-driver of "real data + synthetic data" to ensure closed loops in real scenarios while solving the long-tail data generalization challenge in complex environments?

·First public disclosure of the robot's "Five Layers of Autonomy": Remote Control → Environmental Autonomy → Supervised Control → Task Autonomy → Full Autonomy.

·The logic behind Unitree's open-source community UnifoLM: Open-sourcing only "underlying thinking" instead of directly providing "cool actions" to teach people how to fish.

·Key pain points of large model deployment: Three major bottlenecks including edge-side computing power, hallucinations, and latency. Unitree presents an "edge-cloud collaboration" solution.

·Three pillars of safety and governance: Recoverability, interpretability, and generalization, allowing humans to trust robots with confidence.

·Three pieces of advice for students wanting to enter embodied intelligence: Passion over trend-chasing, mastering both software and hardware, and writing your first line of code by hand.

Whether you are a robot developer, an entrepreneur in the embodied intelligence track, or a professional focusing on humanoid robot deployment, this episode will uncover the technical details, commercialization paths, and future trends of the integration of AI and embodied intelligence. Stay tuned and follow GOSIM for more practical insights on open-source and AI+robotics!