Post by Avery Gump

👻 CVPR 2026: A Fantastic Week of Tech and Innovation! 🚗💨 I had an incredible time sharing our latest research and presenting our paper: "Ghosts in the Point Clouds: De-Glaring LiDAR in the Transient Domain." If you've ever worked with autonomous vehicles or robotics, you know that "blooming" is a critical failure mode for these sensors. Our work explores how operating within the transient domain allows us to recover vital scene information before it is lost. 🔗 Project page: https://lnkd.in/gUXbeYdA 🤝 Huge thanks to my collaborators: Connor Henley Sungjin Cheong Akarsh Prabhakara Mohit Gupta 💡 Key Takeaways A major theme from my time at CVPR 2026 is that we cannot simply wait to encounter long-tail failures on the road; we must proactively prepare for them to avoid safety-critical incidents. To do this, we need to create these edge cases through simulation and train our models on this data, rather than relying solely on the vast amounts of data collected under nominal conditions. The team at Waabi shared some fantastic insights on this challenge during their tutorial. While my own research focuses earlier in the processing pipeline (prior to downstream perception), it was fascinating to see the massive wave of work utilizing foundation models for perception, alongside generative models for world and sensor simulation. Physics-informed sensor simulation will be absolutely key to bridging the real-to-sim gap, which is vital to ensuring these models maintain their performance when they hit the road. 🙌 To everyone I connected with during the week: Thank you for the amazing conversations, presentations, and insights! I hope to see you all next year (or even sooner if you are heading to ICCP 2026)! If you missed the presentation but want to chat about LiDAR physics, transient imaging, or the future of autonomous sensing, feel free to reach out or drop a comment below! 🔍👇 #LiDAR #AutonomousVehicles #Robotics #ComputerVision #AI #AutonomousTrucking #SensorFusion