Algorithms for AI Driving Olympics: compressing information with autoencoders
Autonomous control algorithms for (small) vehicles are a challenging problem. Events like the Driving Olympics are an excellent testbed for new algorithms in cyber-physical domains. New frameworks with AI and ML libraries make it much easier to integrate custom ML solutions with cyber-physical systems without relying on expensive hardware.
One of the critical subsystems for autonomous driving is denoising during the pre-training process. Experimenting with Variational Autoencoders (VAE) and visual domain randomization resulted in efficiently compressing input information into a latent representation.
#variationalautoencoders #drivingolympics #deeplearning #AI #VAE #latentrepresentation
#visualdomainrandomization #VDR
https://www.duckietown.org/research/ai-driving-olympics
https://www.ttic.edu/ripl/assets/publications/censi19.pdf
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.