Delta Intelligence raises nearly $70M Angel++ round for humanoid robot foundation model. The UCLA Ph...
The AMW Read
Introduces a new entrant with a distinct native-3D architecture in the humanoid robot foundation-model segment, updating the player map and challenging the 2D-plus-patching paradigm, with potential segment-level impact if successful.
Delta Intelligence raises nearly $70M Angel++ round for humanoid robot foundation model. The UCLA PhD-founded startup, established in January 2026, has completed its sixth financing round in under six months, totaling nearly 500M yuan (~$70M). Investors include humanoid robot OEMs such as Agibot, UBtech, and Unitree, as well as financial institutions like GL Ventures and Lenovo Capital. The company is developing a native 3D world engine and a full-body loco-manipulation foundation model (HFM) for bipedal humanoid robots, aiming to move beyond 2D visual representations that fail in real-world spatial reasoning.
Why it matters: Delta Intelligence is pursuing one of the hardest open challenges in embodied AI — a general-purpose, real-time 3D foundation model for humanoid locomotion and manipulation — in a capital environment where many early-stage robotics startups are still proving lab-to-deployment viability. The round’s structure (Angel++ with industrial OEM co-investors) signals a strategic pattern we have seen in earlier substrate cycles: hardware incumbents placing small bets on algorithm-first teams to secure future model access, a variation of the acqui-licensing pattern. The company’s explicit rejection of 2D-visual patching in favor of a native-3D approach positions it against all major existing humanoid robot stacks, which rely on 2D vision-language models adapted for robotics. If Delta Intelligence can demonstrate reliable full-body loco-manipulation in industrial settings (energy O&M, outdoor inspection), it could reshape the architecture debate for the Physical AGI substrate.
Expert take: What makes Delta Intelligence noteworthy is not the round size — which is modest by 2026 standards — but the speed of capital validation: six rounds in six months from a mix of OEMs, financial VCs, and semiconductor investors. This matches the fastest-ARR-ramp pattern we have observed in previous dominant-substrate plays, where trust in the founding team (three UCLA PhDs with Google Robotics, NVIDIA Research, and DARPA project experience) is traded for compressed diligence cycles. The hard test will be whether their native-3D approach can achieve the data volume required for robust generalization, since their self-developed full-body panoramic capture rig is still in pilot production. If they fail to solve the data-closed-loop at scale, the ‘native 3D vs. 2D-plus-patching’ debate will remain unresolved.
#DeltaIntelligence #HumanoidRobot #EmbodiedAI #FoundationModel #RoboticsFunding #PhysicalAGI