This Startup Wants to Build Self-Driving Car Software—Super Fast (www.wired.com)

HyprLabs, a startup founded by former Zoox co-founder Tim Kentley-Klay, has emerged with an innovative approach to developing self-driving car software, showcased through two modified Tesla Model 3s equipped with extra cameras and a compact supercomputer. Operating in San Francisco, the startup employs a novel training technique called "run-time learning," which enables its vehicles to learn in real-time with minimal data input. Unlike traditional approaches that rely heavily on extensive pre-collected datasets, HyprLabs aims to efficiently train its systems by capturing only novel data while on the road. To date, its Teslas have logged 4,000 hours of driving data, significantly less than competitors like Waymo, yet the company emphasizes its ability to improve safety and performance with less computational work. This announcement holds considerable significance for the AI/ML community as it represents a potential paradigm shift in autonomous vehicle development. By blending the strengths of camera-only and multi-sensor approaches, HyprLabs proposes a more agile and cost-effective solution to training autonomous driving systems. While the startup acknowledges it is not yet ready for public deployment, its progress could pave the way for new categories of robotic systems. With plans to introduce its unique robot in the next year, HyprLabs is poised to challenge existing norms in the autonomous vehicle landscape, potentially accelerating advancements in the field.