bloxyen in  
Software Engineer  

Google’s new DeepMind Maps algorithm improves route suggestions by 24%

Google has developed an AI algorithm to refine route suggestions on Google Maps, personalizing it based on user data and behavior, allegedly improving the accuracy on an average by 16-24 percent.

To stay on top of the latest advancements in AI, look here first.

Personalized Route Suggestions through AI

  • The AI model comprises 360 million parameters, using real-time data from Maps users to influence diverse factors including travel time, road conditions, tolls, and personal preferences to suggest routes.
  • This technology is grounded on "inverse reinforcement learning" (IRL), specifically a new IRL algorithm - "Receding Horizon Inverse Planning (RHIP)".

The Power of RHIP and AI in Maps

  • Google and Deepmind jointly worked to develop RHIP, using complex stochastic models in immediate vicinity areas, but switching to simpler deterministic methods for distant areas for power conservation.
  • The AI improves route suggestions for both driving and two-wheeled vehicles by learning from Maps users' movements and behaviors over time.
  • Google states that this is the largest application of inverse reinforcement learning for route planning to date.

Implementation and User Testing

  • Google has applied the algorithm to Maps data globally, but extensive user testing is needed to confirm if the technique consistently produces better routes.
  • Previous attempts at using AI systems for route planning on a large scale have often failed due to the complexity of road networks.


P.S. If you like this kind of analysis, I write a free newsletter that tracks the most relevant news and research in AI. Professionals from Google, Meta, and OpenAI are already reading it.

Supercharged AI

Supercharged AI

Stay on top of the latest AI news, research, and tools with our daily 2-min email. Join 1000+ readers from Google, Meta, OpenAI, and more.