HashMap: The next generation Google Maps using simulation-based traffic prediction
Simulation-based digital twin for complex real-world traffic modeling to enable accurate prediction in ‘impossible to model’ traffic scenarios for critical decision making.
Berkeley, CA, November 2020 – Using the newly created Hash.AI simulation tool, 4 students from the University of California, Berkeley, have come up with a traffic simulation of delivery-cars in the city of Berkeley, CA. It helps predict the efficiency of delivery services given partner stores in a city.
Details – Real world traffic is very complex and dynamic. Jaywalkers, bikers, truckers, cars, travelers, varying weather, holidays, rush hour, accidents, and autonomous vehicles are just some of the features and agents that play a key role in determining traffic patterns. However, much of these smaller details are unaccounted for in what mapping apps claim to be real-time, real-world analysis, but these smaller details can have a significant and cascading effect on traffic congestion. Prediction of such random processes, like when and where people will go shopping for groceries, with real-time implementation is an intractable problem. Simulation is the next-best method to approximate a prediction on how complex interacting agents will behave given large and varying inputs. Using HASH.AI, a startup that is building an end-to-end solution for simulation-driven decision making, we have developed a small-scale version of the city of Berkeley to efficiently visualize how every agent interacts and make decisions about the future of the city’s traffic policies.
“This is the first simulation that measures the impact of the different road conditions on the service time of delivery businesses.” said Malo Le Magueresse, a member of the team that led the project. “Its impact on the sector could be huge, and it could potentially help companies shift their strategy at an unprecedented granularity: within each city or even neighborhood!”
HASH is an open platform for simulating anything. Components in HASH are mapped to extensible open schemas that describe the world. These can be combined to quickly create accurate digital-twins of our complex real-world.
The proof – The model created by the team at Berkeley simulates the demand of deliveries based off of store locations scrapped from Yelp and randomly generated home locations with family sizes pulled from the census data. Utilizing the power behind HASH.AI, the team was able to simulate the transactions of the purchase of goods along with generating data of potential costs of managing such a system. The ease of scalability of the model allows for simulations to be generated for different cities quickly due to the usage of smart management of code files. Creation of more agents is relatively easy as the basic framework has been developed and definition of more behaviors is simple to add to the powerful HASH.AI system that it is running off of. The sample presented above can easily be scaled up to larger projects due to the nature of modeling agents in the HASH.AI ecosystem.
The takeaways – Simulation driven real-time decision making for traffic congestion and navigation routing is now available. For road users, we offer more accurate predictions of traffic conditions. For delivery platforms, we anticipate demand, efficiently route drivers, and measure delivery time and customer satisfaction. The possibilities to disrupt the industry are endless, and we look forward to a future where traffic simulation can bring about positive societal change.
Malo Le Magueresse