Nuro Raises $203M for AI-First Autonomous Delivery

Autonomous delivery leader Nuro secures $203M to scale its AI-driven self-driving tech and expand commercial partnerships with major retailers.

Boston Dynamics and the Toyota Research Institute (TRI) are fundamentally shifting how humanoid robots learn, moving away from painstaking line-by-line code and toward a more intelligent, scalable approach. They've successfully implemented large behavior models, a specialized form of AI, to train the iconic Atlas robot to perform complex, multi-step tasks it has never encountered before. This isn't just about better movement; it's about giving the robot a form of semantic understanding, allowing it to interpret high-level commands like "pack these tools into the toolbox" and then autonomously generate the sequence of actions to make it happen. A senior TRI engineer stated, "We're moving from scripting every motion to creating a generalized framework for physical reasoning, which is a critical step toward practical utility in unstructured environments."

Under the hood, this tech is seriously next-level. The system leverages a library of learned skills, from dynamic walking and grasping to force-based manipulation, which the behavior model can chain together in novel ways based on the task. When given a new objective, Atlas's onboard AI runs through a rapid simulation cycle, predicting the outcomes of various actions using deep reinforcement learning before executing the most efficient plan. This allows it to handle uncertainty and make real-time adjustments, like precisely placing a heavy car strut in a specific orientation or recovering balance when an object weighs more than anticipated, all without a human in the loop writing new control algorithms for each scenario.

The implications for this are absolutely massive for real-world automation in industries like manufacturing and logistics. This breakthrough directly addresses the biggest bottleneck in robotics: the inability to adapt. By using large behavior models, Boston Dynamics and TRI are creating robots that can be tasked through intuitive language and that can generalize their learning from one factory floor to another without costly reprogramming. This isn't just a lab demo; it's a clear roadmap toward a future where humanoid robots can tackle dull, dirty, and dangerous jobs, adapting on the fly to new challenges and finally fulfilling the promise of flexible automation. As one project lead put it, "The goal is a robot that understands the intent behind the instruction, not just the command itself."