September 2026 Summaries

Encord will host its inaugural Long Horizon Physical AI Summit at Fort Mason in San Francisco on October 20–21, 2026, bringing together more than 60 speakers from organizations including Waymo, Google DeepMind, Boston Dynamics, Amazon Robotics, and Skild AI. The two-day event focuses on robotics on the first day and autonomy on the second, examining subjects such as humanoid manipulation, robot learning data pipelines, world models, large behavior models, autonomous-vehicle deployment, fleet intelligence, navigation, and autonomy in contested environments. Featured speakers include Skild AI CEO Deepak Pathak, Wayve CEO Alex Kendall, Google X and Waymo founder Sebastian Thrun, and executives and researchers from leading robotics, automotive, defense, and AI companies. The program also includes technical workshops on automotive-plant instrumentation, trainable robots, hand dexterity, perception-model evaluation, and sensor-fusion data enrichment, with several major technology and business media outlets expected to attend.

Physical AI, spanning robotics applications such as manufacturing, logistics, surgery, agriculture, and autonomous vehicles, may become a major growth market, with Nvidia projecting related revenue to rise from $10 billion to $100 billion over the next decade. The central challenge, however, is argued to be data diversity rather than data volume: unlike digital AI fields that rely on broadly uniform text and screen-based interactions supported by internet-scale data, physical systems must learn across varied sensors, robot bodies, materials, environments, tasks, and safety-critical failure conditions. Research identifies this “embodiment heterogeneity” and the high cost of collecting real-world demonstrations as significant obstacles, while existing cross-robot datasets remain limited relative to the range of conditions needed for reliable deployment. The gap between impressive demonstrations and scalable commercial operations may reflect these data constraints, particularly when models must transfer between facilities, hardware types, and tasks. Encord presents its multimodal data ingestion, annotation, quality control, and curation platform as infrastructure intended to help physical AI teams manage and reuse heterogeneous data across sensor types, robot embodiments, and industry verticals.