Excellent synthesis of where the field stands. The Chain-of-Action framing is probly the most underrated development here, basically solving the interpretability problem that's plagued end-to-end robotics for years. Having those intermediate plan tokens means you can actully debug why a system fails instead of just watching it fail mysteriously. The Wayve 90-city deployement with 62% unseen cities is wild evidence for generaliztion. Reminds me how much real-world variance matters compared to sim benchmarks.
Great coverage on the VLAM topic that aligns well with the progress on overall "Spatial Intelligence" space. Basis like NeRF: Neural Radiance Fields paper - that readers might find useful -
Wayve is probably your best investment yet has a VC.
Excellent synthesis of where the field stands. The Chain-of-Action framing is probly the most underrated development here, basically solving the interpretability problem that's plagued end-to-end robotics for years. Having those intermediate plan tokens means you can actully debug why a system fails instead of just watching it fail mysteriously. The Wayve 90-city deployement with 62% unseen cities is wild evidence for generaliztion. Reminds me how much real-world variance matters compared to sim benchmarks.
Great coverage on the VLAM topic that aligns well with the progress on overall "Spatial Intelligence" space. Basis like NeRF: Neural Radiance Fields paper - that readers might find useful -
July 2025 - Emerging Technology — Spatial Intelligence: Unlocking 3D Understanding in AI - https://sanjeevarora.substack.com/p/emerging-technology-spatial-intelligence-unlocking-3d-understanding-in-ai