A deep-dive into the Wayformer architecture — how factorized attention over agents and scene context achieves state-of-the-art trajectory prediction while staying practical for production AV stacks.

How switching from Cartesian to polar coordinates for LiDAR point cloud processing cuts latency by 9× without hurting accuracy — and the non-obvious trade-offs involved.

How NeRF-based scene representations are replacing hand-crafted 3-D asset pipelines in AV simulation — what we built at Rivian, and where the method still breaks down.