🤖 XPeng’s Humanoid Spin-Off Dogotix Raises $900M, Valued at $6.3B

• ●Dogotix, the humanoid robotics unit spun out of XPeng Motors, closed a $900M funding round — pushing its valuation to $6.3 billion before a single mass-market robot ships.
• ●The raise signals that XPeng is treating humanoids as a standalone platform business, not a feature — separating the capital stack from its EV operations to accelerate scale manufacturing.
• ●Watch for production volume targets and factory announcements: at $6.3B, investors are pricing in a credible path to thousands of units, not lab demos.
• 🔍 Read More →
• What matters: A $6.3B valuation for a pre-production humanoid unit means the market is betting on XPeng’s manufacturing DNA, not just its AI roadmap.

🧪 Technology, Research & Innovation

🧠 NVIDIA Cosmos-H-Dreams Brings Real-Time Generative Physics to Surgical Robotics Simulation

• ●Cosmos-H-Dreams is a new generative simulation framework from NVIDIA that enables real-time interactive physics for surgical robotics — replacing costly animal and cadaver lab experiments.
• ●Classical simulators fail to reproduce photorealistic appearance and deformable-tissue dynamics simultaneously; Cosmos-H-Dreams addresses both in a single real-time pipeline, a technically non-trivial combination.
• ●If validated on physical hardware, this could compress surgical robot training cycles from months of wet-lab work to days of synthetic data generation.
• 🔍 Read More →
• What matters: Real-time deformable-tissue simulation at photorealistic fidelity is the missing link between surgical AI research and deployable robot-assisted procedures.

🤖 NeurRAFT Tackles Collision-Free Robot Motion Planning in Cluttered Environments

• ●NeurRAFT is a generative motion planning framework that uses anchor-level flow matching and clearance-aware preference tuning to generate collision-free trajectories directly from raw sensor observations.
• ●Unlike classical planners that require privileged geometric models, NeurRAFT is end-to-end neural — the clearance-aware tuning step is the key differentiator for dense, cluttered scenes where margin errors are costly.
• ●The approach targets a persistent gap in neural planners: safe operation in unstructured environments, which is the primary blocker for deploying manipulation robots outside controlled settings.
• 🔍 Read More →
• What matters: Clearance-aware preference tuning gives neural planners a principled safety margin — the ingredient that has kept end-to-end approaches out of real-world manipulation deployments.

🚀 Product, Hardware & Model Launches

🤖 NVIDIA Jetson Orin Nano 2 Targets Edge AI for Drones and Robots

• ●NVIDIA launched the Jetson Orin Nano 2, a new edge compute module designed to bring physical AI inference to cost-sensitive drone and robotics platforms.
• ●The Orin Nano 2 extends the Jetson Orin family’s power-efficiency profile to smaller form factors, enabling on-device perception and control loops that previously required larger, more power-hungry modules.
• ●For robotics developers, the key question is whether the Nano 2’s performance-per-watt unlocks new deployment categories — particularly autonomous drones and lightweight mobile manipulators — where thermal and battery budgets are hard constraints.
• 🔍 Read More →
• What matters: Pushing physical AI inference to smaller, cheaper edge modules is how NVIDIA converts its robotics software ecosystem into hardware volume at scale.

🚗 Waymo’s Custom AV Chip Revealed: 160 TOPS Real-World Performance Behind the 1000+ TOPS Headline

• ●Details of Waymo’s self-developed autonomous driving chip have surfaced, showing a headline spec above 1000 TOPS — but the operationally relevant figure is 160 TOPS of real-world, sustained inference performance.
• ●The gap between peak and sustained TOPS is the critical engineering story: real-world AV workloads involve continuous multi-sensor fusion under thermal constraints, where peak figures are largely irrelevant to system design.
• ●A custom silicon strategy lets Waymo co-optimize hardware and software stack — reducing per-mile compute cost and reducing dependency on third-party suppliers as it scales its commercial fleet.
• 🔍 Read More →
• What matters: 160 TOPS sustained is the number that actually determines fleet economics — Waymo’s custom silicon bet is fundamentally a cost-per-mile play, not a benchmark race.

💰 Business, Startups & Investment

🧠 General Intuition Raises at $6B Valuation as Valor and Point72 Back Spatial Foundation Model

• ●General Intuition is in talks to raise at a $6B pre-money valuation, with Valor Ventures, Point72 Ventures, and Seven Seven Six among new investors backing its spatial foundation model for AI agents.
• ●The startup’s core thesis is a generalized model that trains AI agents to move through space and time — positioning it as infrastructure for robotics rather than a single-application product.
• ●A $6B pre-money valuation for a foundation model company without a shipping robot product reflects investor conviction that spatial reasoning will be the next platform layer — and that the window to own it is now.
• 🔍 Read More →
• What matters: Investors are pricing spatial foundation models like platform infrastructure — the same bet made on LLMs in 2021, now applied to physical world reasoning.

🚗 Gatik Raises $200M After PepsiCo Deal, Doubling Down on Middle-Mile Autonomous Trucking

• ●Self-driving truck startup Gatik closed a $200M funding round, with the raise coming on the heels of a commercial deal with PepsiCo — providing a revenue anchor that most AV trucking competitors lack.
• ●Gatik’s middle-mile, fixed-route strategy deliberately constrains the operational design domain, which has allowed it to reach commercial deployment faster than open-highway autonomy players.
• ●The PepsiCo deal matters as a template: enterprise CPG contracts give AV trucking startups the predictable utilization rates needed to justify fleet expansion and attract follow-on capital.
• 🔍 Read More →
• What matters: Gatik’s PepsiCo anchor proves that constrained-ODD autonomy with enterprise contracts is a fundable, scalable business model — not a consolation prize for companies that couldn’t crack open-road autonomy.

⚡Humanoid Capital::Two companies — Dogotix and General Intuition — are raising at $6B+ valuations simultaneously, signaling that the humanoid and spatial AI funding cycle has entered a new magnitude.

⚡Custom Silicon::Waymo’s 160 TOPS sustained performance figure reframes the AV chip conversation — real-world efficiency, not peak benchmarks, is what determines fleet unit economics.

⚡Simulation Fidelity::Cosmos-H-Dreams targets surgical robotics, but the real prize is a generalized real-time deformable-physics simulator — whichever team cracks that wins the synthetic data race for contact-rich manipulation.

⚡Constrained ODD as Strategy::Gatik’s $200M raise on the back of a PepsiCo deal validates fixed-route middle-mile autonomy as a durable commercial model, not a stepping stone to something bigger.

⚡Edge Compute Democratization::If Jetson Orin Nano 2 delivers meaningful physical AI inference at drone-class power budgets, the next debate will be whether NVIDIA’s edge ecosystem or open alternatives win the sub-$500 robotics compute market.