Cruise teaches robotic vehicles to navigate unfamiliar situations

Cruise teaches robotic vehicles to navigate unfamiliar situations

September 15, 2020 0 By autotimesnews

GM-owned California-based startup Cruise has unveiled its new development – a tool for training robotic vehicles to act in unusual situations on the road. For example, the sudden appearance of a cyclist or a child on the highway, or a gross violation of traffic rules by other vehicles.

One of the challenges facing autonomous vehicle developers is to teach robotic vehicles to predict the intentions of other road users. People don’t always follow the rules, and even then they often make mistakes. According to the US National Traffic Safety Administration, 94% of serious accidents are caused by mistakes or risky decisions by drivers.

That’s why the startup Cruise, a subsidiary of General Motors, has developed the Continuous Learning Machine, writes VentureBeat. Thanks to active learning technology, it automatically detects errors that make models, and only those scenarios in which there is a significant difference between prediction and reality are included in the Cruise machine training database. This allows for more concentrated data collection and minimizes overly light scenarios.

The Continuous Learning Machine labels data autonomously using a “basic truth” model. Basically, the framework assumes what a person or machine can do in the future, and compares it to what they actually do. The final stage is training a new model, testing it and testing it in road conditions.

According to the developers, the tool allows you to make extremely accurate predictions for a number of rare situations that can occur in real life. In particular, the 180-degree turns that Cruise vehicles do not see very often on the streets, or sudden lane changes when the car tries to avoid slowing down.

In February, California authorities granted Cruise autonomous vehicles permission to conduct pilot tests on state roads. If they are successful, the company will be able to bring 2,500 of its robotic vehicles onto the road at once.