Self-driving company Waabi is changing the game using a generative AI model to forecast vehicle movement. They’ve unveiled a new system called Copilot4D, which was trained using data from Light Detection and Ranging (LiDAR) sensors. When a driver recklessly merges onto a highway at high speed, the model predicts how surrounding vehicles will move, generating a LiDAR representation of 5 to 10 seconds into the future. This allows the driving software to react accordingly. A more advanced version is already being used in Waabi’s fleet of autonomous trucks in Texas.
What’s Happening & Why This Matters
Autonomous driving has traditionally depended on machine learning to plan routes and detect objects, but now innovators like Waabi are harnessing generative AI models to take things further. Their method breaks LiDAR data into chunks based on training data, then predicts how the data will move, allowing it to produce predictions 5-10 seconds ahead.
Waabi’s mission is to design systems that learn from data rather than relying on pre-programmed reactions to different situations. They’re not alone in this approach, with competitors like Ghost and Wayve also classifying themselves as “AI-first.” A key advantage to this approach is that it may require fewer hours of road-testing self-driving cars than traditional methods. Waabi is the first to build a generative model for LiDAR rather than cameras.
Waabi’s model has limitations in terms of how far into the future it can estimate, but it excels at projecting 5 to 10 seconds ahead for most driving decisions. The company is still considering whether to make this innovation open-source, a choice that would undoubtedly influence competition and the future of self-driving. In the meantime, Waabi is determined to maintain a balance between openness and competitive advantage.
TF Summary: What’s Next
Waabi is embracing a generative model to revolutionize autonomous driving, showcasing the possibilities of AI in advancing vehicle safety and performance. This shift could have far-reaching effects in the field of autonomous driving and generative AI. The decision of whether to make this model open-source remains an important factor that could shape future developments in this fast-evolving industry.
