We have developed a new breakthrough technology that integrates with any autonomous driving software at both the level of stabilization and behaviour planning.
At the highest level, we predict for each safety relevant object, where and when their trajectory will intersect with the ego trajectory (that would be a collision). For each trajectory, our technology estimates the residual risk, so the autonomous driving system can choose the safest ego trajectory.
We don’t miss anything
But without billions of miles worth of training data, what if we encounter something unknown? Like the unknown behaviour of that skater.
No problem: Our technology can instantly detect all the things and behaviours that it has not been trained for.
But does that not cause way too many disengagements?
No, we automatically adapt our perception model to the unknown things and behavours. Over the entire fleet and without any labelled training data.
Ready for whatever you want
Better AI is safe and also super fast. You can run it on an ECU as well as on a small embedded hardware like an AI processing unit on a camera.
For most of our demos, we run Better AI inside of ROS 2.
Sometimes, we run it on an iPhone.