Building on the trove of data it released in 2019, the Alphabet company is calling this latest batch “the largest interactive dataset yet released for research into behavior prediction and motion forecasting for autonomous driving.”
This “motion dataset” includes objects like cars and people and their trajectories, as captured by the sensors on Waymo’s vehicles. Along with the objects, 3D maps and geographic details are provided along with each segment to give researchers context for their prediction modeling. In total, Waymo says it is releasing 570 hours of “unique data.” The data are unique in that they were mined specifically for their suitability for research purposes, such as scenarios including “cyclists and vehicles sharing the roadway, cars quickly passing through a busy junction, or groups of pedestrians clustering on the sidewalk.” A variety of road types in different urban environments are also featured, including San Francisco, Phoenix, Mountain View, Los Angeles, Detroit, and Seattle.
Along with this release of data, Waymo is also launching an open dataset challenge to research teams across four areas: motion prediction, interaction prediction, real-time 3D detection, and real-time 2D detection. The winner of each challenge will receive $15,000, with second-place teams receiving $5,000 and third place $2,000. This would not be the first time Waymo has released data from its AVs to the broader research community. Last year, the company published 6.1 million miles of driving data, including 18 crashes and 29 near-miss collisions.
This announcement comes hot on the heels after another recent Waymo development, where the company simulated dozens of real-world fatal crashes that took place in Arizona over nearly a decade. The simulations reconstructed the scenarios leading up to the crashes for Waymo’s vehicles to see how they would have reacted if they were put in that same situation. According to Waymo, the results showed that their autonomous vehicles would have avoided or mitigated 88 out of 91 total simulations. Moreover, for the crashes that were mitigated, Waymo’s vehicles would have reduced the likelihood of serious injury by a factor of 1.3 to 15 times.
Access to data from Waymos autonomous vehicles can be of great benefit for the research community. Many researchers have experienced how difficult it can be to get data even when working together with OEMs or operators in R&D projects. Data sharing is often a major limiting factor to the effectiveness of public research in the AV field.
Data usability is another issue. In this case, it seems that Waymo is making more extensive data sets and context available. The inclusion of not only objects such as cars and people, but of their trajectories, corresponding 3D maps and geographic details as context for prediction modeling shows that Waymo has a good understanding of researcher needs around data usability. It also seems promising that a variety of road types in different urban environments are included and that the data is mined specifically for research purposes – this could significantly reduce the amount of time researchers devote to sifting through data. Yet due to the lack of data standards in this field, it could still be time-consuming for researchers to make use of the open-source datasets for a myriad of other reasons.
It is interesting that Waymo has decided to involve the broader research community to help it solve certain aspects of the AV challenge, like behaviour prediction, even though the broader research community is nowhere near as well equipped and resourced as Silicon Valley backed operations like themselves and others like Zoox. This indicates that they do not necessarily see the AV race as a zero-sum game, but can see benefits in leveraging the infrastructure of expertise that exists in the public research community even if knowledge gained there could also help their competitors.
Written by Cilli Sobiech,
RISE Mobility & Systems (Mobilitet i transformation)