The firm, based in California and with a subsidiary in Shanghai, says it believes its expanded dataset “to be the largest interactive dataset yet released for research into behaviour prediction and motion forecasting for autonomous driving.”

Waymo’s Chief Technology Officer Dmitri Dolgov says interesting motion data is hard for researchers to gather themselves. “Most day-to-day driving is uneventful – which makes for uninformative data when you are building a system to predict what could happen on the road in unusual situations. As a result, existing datasets often have a limited number of interesting interactions.”

But the expanded dataset is mined specifically to include interesting examples: cyclists and vehicles sharing the roadway, cars quickly passing through a busy junction, or groups of pedestrians clustering on the sidewalk.

Sample from a Waymo contest clip: could you predict the future interactions at this busy San Francisco junction?

Researchers, students and citizen scientists are invited to use the dataset to take part in Waymo’s Challenge, the toughest of which requires contestants to predict the positions of two interacting agents for eight seconds into the future, based on just one second of motion data.

Other challenges include producing a set of 2D boxes for objects in a scene as quickly as possible, as measured by a Nvidia Tesla graphics card: the fastest will win.

Waymo’s put up a US$15,000 (HK$120,000) top prize for each of the four challenges.

The firm has previously said it aims to build “the world’s safest driver”. Last year it released a detailed report on its safety experience with results from 10 million km of automated driving and 100,000 km of fully automated “no human” driving: Waymo’s vehicles had 18 crashes with a pedestrian, cyclist, driver, or other object, while experiencing 29 “disengagements”, where a safety driver took control of the wheel to prevent a crash.

“This data represents over 500 years of driving for the average licensed US driver,” said Waymo at the time.