The Recipe For Success
Building Self-Driving Cars Without Big Data?
Founding Story
Am not going to lie to you. We’re a startup company founded in Sept 2023. Doing self-driving cars in terrace terrains in Agriculture; the hardest terrain in the world. Worse yet, we also lack big-data and huge compute. So, what makes us unique?
Less is More
Is it Necessary?
Big data isn’t necessary. As of now, the world has gathered sufficient amounts of real-world driving data and stored them in nu-scenes dataset. What’s lacking in the world for self-driving is correct methods to tackle.
– out-of-distribution data
– correct recipe for making a self-driving car
What is the correct recipe?
Imagine if we could predict 5 seconds into the future with 100% accuracy. But we must take 6 seconds to predict it. Is it useful for online predictions ?
For robotics, we mainly do online predictions rather than offline, therefore one aspect of the correct recipe should be a compromise of speed and accuracy in each pipeline to perform the task with 100% accuracy.
Handling out-of-distribution without BIG-DATA?
Data augmentation of course! It seems simple. Research has shown state-of-the-art AI’s that use real-data failed to understand the context that is easily simulated with simple data-augmentation.
We in GrootCompute tries to retain the basics while combining generative-AI concepts to tackle these issues in self-driving cars.
