Robotics News: Baidu Robotaxi Mass Failure, NVIDIA Physical AI Models, Japan Robot Strategy — April 7, 2026

April 7th 2026

The humanoid race just crossed its first real production milestone -- and China is setting the pace.

It was National Robotics Week and the industry delivered more than just celebrations. AGIBOT rolled out its 10,000th humanoid robot in Shanghai, Japan declared physical AI a matter of national survival with $6.3 billion on the table, and NVIDIA dropped new open models to cement itself as the platform layer beneath every robot on Earth. The gap between prototype and production line has never closed this fast.

Watch & Listen First

Shawn Ryan Meets a Humanoid Robot -- Figure AI CEO Brett Adcock demos Figure 03 live | Shawn Ryan Show #292 Balance tests, gait demos, wireless charging, and Adcock's first public comments on severing ties with OpenAI. The robot assembles off the line every 90 minutes.

Embodied AI 101 -- Daily research briefings on physical intelligence | Spotify New dedicated podcast covering embodied AI papers and industry developments daily. Spun off from Better Call Shao in April with fresh episodes on world models and sim-to-real transfer.

IEEE Spectrum Video Friday: Humanoid Dancing, Robot Learning, More | IEEE Spectrum This week's roundup includes Agility teaching Digit new dance moves through automated training, plus clips from labs worldwide pushing locomotion and manipulation.

Key Takeaways

  • 10,000 humanoid units is no longer a forecast -- it is a fact. AGIBOT hit the milestone on March 30, compressing from 5,000 to 10,000 units in just three months. Deployed across logistics, retail, hospitality, and education in 20+ countries, this is the production curve that turns skeptics into order sheets.

  • Japan is treating physical AI as a national survival strategy. METI wants 30% of the global physical AI market by 2040. Mujin already runs autonomous picking across 100+ warehouse sites. With the working-age population in structural decline, Japan is proving that the strongest use case for robots is filling jobs nobody is left to take.

  • NVIDIA is building the Android of robotics, and the ecosystem is locking in. The new GR00T N1.6 vision-language-action model, Cosmos Predict 2.5 world models, and Isaac Lab-Arena evaluation tools form a full sim-to-real pipeline. Partners from ABB to FANUC to Figure are building on it.

  • Competitions are becoming real capability benchmarks. Beijing's humanoid half-marathon (April 19) will field 300+ robots with a new autonomous navigation group. The URKL combat league opened global registration for full-contact humanoid fighting. These are not stunts -- they expose the gap between lab demos and field performance.

  • Embodied AI is graduating from papers to developer conferences. X Square Robot's inaugural EAIDC brought the first large-scale gathering focused on embodied AI deployment, with real-robot task execution and end-to-end evaluation workflows.

    • The Big Story

    AGIBOT Rolls Out Its 10,000th Humanoid Robot, Signaling Industry Shift to Mass Deployment · March 30, 2026 · Robotics & Automation News

    → The Shanghai-based company compressed its production timeline dramatically -- the first 1,000 units took nearly two years, but the leap from 5,000 to 10,000 happened in a single quarter. Deployments span Europe, North America, Japan, South Korea, Southeast Asia, and the Middle East. This is not just a Chinese milestone; it is the moment the global humanoid industry has to reckon with a competitor that has actual scale economics, not just valuation economics.

    Also This Week

    Japan Commits $6.3B to Physical AI as METI Targets 30% of Global Market by 2040 · April 5, 2026 · TechCrunch → Mujin's software-first approach -- making existing industrial robots autonomous -- is the sleeper strategy while everyone else builds humanoids from scratch.

    NVIDIA Releases GR00T N1.6 and Cosmos Predict 2.5 for Physical AI · April 2026 · NVIDIA Newsroom → Open VLA models plus the new Jetson T4000 with 4x energy efficiency make NVIDIA's full sim-to-real stack increasingly hard to compete with.

    Beijing Humanoid Half-Marathon Confirms 300+ Robots for April 19 Race · March 2026 · Global Times → The new autonomous navigation group -- 38% of entrants -- is the real signal: teams are betting on untethered, self-directed locomotion over teleoperation.

    URKL Humanoid Combat League Opens Global Registration · April 3, 2026 · GlobeNewsWire → EngineAI's T800 platform -- 1.85m, 85kg, 41 degrees of freedom -- becomes the standardized testbed; the $1.45M gold championship belt ensures teams take it seriously.

    AI-Enhanced Robotics Gains Traction in Pharma Manufacturing · April 7, 2026 · GlobeNewsWire → TechForce Robotics integrates real-time AI deviation detection into pharmaceutical production lines -- contamination control is the wedge into regulated industries.

    From the Lab

    Building Generalist Humanoid Capabilities with GR00T N1.6 Using Sim-to-Real Workflow · NVIDIA Technical Blog

    → The sim-to-real pipeline combining whole-body reinforcement learning in Isaac Lab with synthetic navigation data achieves zero-shot cross-embodiment transfer -- a policy trained on one humanoid body can run on a different one with minimal finetuning. This is a foundational step toward truly general-purpose robot software, where the model matters more than the metal.

    Worth Reading

  • National Robotics Week: Latest Physical AI Breakthroughs -- NVIDIA's roundup covers Aigen's solar-powered farm bots, Maximo's 100MW robotic solar installation, and the full breadth of physical AI going mainstream.
  • X Square Robot Hosts Inaugural EAIDC 2026 -- First developer conference built entirely around deploying embodied AI in production -- hackathons, real-robot evaluations, and commercialization playbooks.
  • Why 2026 Will Bring a Reckoning for Warehouse Robotics -- The warehouse automation market approaches $30B, but expect the first major failures and mergers alongside the growth.

    • The robots are no longer just walking. They are shipping at scale, running half-marathons, and fighting each other in organized leagues. The question has shifted from "can they work?" to "who controls the stack?" This week made the answer a little clearer.