Trajectory raises $50M to advance AI visual understanding for robotics
In brief
- Trajectory, founded by Google DeepMind and Apple researchers, raises $50M in seed funding.
- Multimodal AI systems enable real-time visual understanding for robotics and autonomous vehicles.
- Applications include manufacturing defect detection and continuous physical system learning.
Founding team brings deep AI expertise
Andrew Dai, who spent more than 14 years at Google DeepMind leading data and pre-training efforts for the Gemini model family, is a founder. Yinfei Yang, formerly the chief research scientist at Apple, co-founded the company alongside Seth Neel, drawn from Harvard's AI research community. The trio combines industry-scale AI development with academic rigor.
Addressing AI's visual blind spot
Current large language and multimodal models excel at text and reasoning tasks but struggle with physical understanding. "Your AI assistant can write a legal brief but struggles to tell you whether a shelf is about to fall over," the founders noted. This gap matters for any system that must interact with the physical world.
Trajectory's core strategy involves applying rapid iteration cycles to train AI systems that learn continuously across industries, rather than waiting for massive periodic retraining. The approach mirrors how agile software development accelerated product iteration in the 2010s, but applied to AI model improvement.
Near-term targets
The immediate applications Trajectory is targeting include robotics and physical systems — warehouse robots that need to navigate cluttered environments, autonomous vehicles parsing complex street scenes, and manufacturing systems detecting defects in real time. Each domain generates visual data continuously; Trajectory's framework aims to let models learn from that data stream without waiting months for the next full training cycle.
Coverage of Trajectory's funding efforts first surfaced in The Information in January 2026, with Bloomberg following up in April.
