Motion forecasting – i.e., accurately predicting the behavior of other drivers and road users – is one of the biggest challenges faced by researchers working in the field of autonomous driving. And yet, this is a problem that researchers must overcome because it has enormous safety implications.
Crash avoidance is directly proportional to an intelligent driver’s ability to answer questions like:
- Is that pedestrian trying to cross the street?
- Is that car parallel parked, or about to pull into my lane?
- Will that speeding vehicle stop at the stop sign?
And so, to benefit the driverless research community, Waymo (formerly the Google self-driving project) has just added a new high-quality dataset containing more than 570 hours of unique motion data to its open data repository – Waymo Open Dataset.
Waymo calls this motion dataset “interesting”. It’s interesting because unlike day-to-day, uneventful driving, this data contains object trajectories and corresponding 3D maps for over 100,000 segments, each 20 seconds long, mined for events like:
- Cyclists and vehicles sharing the roadway
- Cars quickly passing through a busy junction, or
- Groups of pedestrians clustering on the sidewalk
According to Drago Anguelov, Distinguished Research Scientist, Waymo, “The Waymo Open Dataset is one of the most geographically varied motion datasets yet released, featuring a wide variety of road types and driving conditions captured around the clock in different urban environments, including San Francisco, Phoenix, Mountain View, Los Angeles, Detroit and Seattle, to encourage models that can better generalize to new driving environments.”
Along with the dataset, the company has also announced new open data challenges to encourage research work on perception and behavior prediction. The challenges focus on areas like motion prediction, interaction prediction, and real-time 2D and 3D detection. The winning team walks home with a $15,000 cash award, while the second- and third-place teams receive $5,000 and $2,000 respectively.