Waymo has released a simulator to train more realistic AI agents

Here's What We Know

In conventional simulators, the behaviour of agents, such as pedestrians, drivers, traffic lights, is pre-set and does not always match reality. Waymax utilises the extensive data set collected by Waymo vehicles while observing real road users. This allows the simulation to model a much more natural reaction of agents to the actions of each other and the unmanned cars.

Benefits of the new simulator include: faster development due to simplified graphics, the ability to focus on behavioural patterns rather than appearance details, and the potential to create safer autonomous driving systems.

Waymax is already available for download on GitHub, although it can't be used commercially yet. Waymo plans to hold task competitions for researchers based on the new simulator.

Overall, the company aims to provide effective tools and data to accelerate the development of unmanned vehicle technology. Such simulators play an important role in testing and improving autonomous driving systems.

Source: Waymo