Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

What-If Motion Prediction for Autonomous Driving

About

Forecasting the long-term future motion of road actors is a core challenge to the deployment of safe autonomous vehicles (AVs). Viable solutions must account for both the static geometric context, such as road lanes, and dynamic social interactions arising from multiple actors. While recent deep architectures have achieved state-of-the-art performance on distance-based forecasting metrics, these approaches produce forecasts that are predicted without regard to the AV's intended motion plan. In contrast, we propose a recurrent graph-based attentional approach with interpretable geometric (actor-lane) and social (actor-actor) relationships that supports the injection of counterfactual geometric goals and social contexts. Our model can produce diverse predictions conditioned on hypothetical or "what-if" road lanes and multi-actor interactions. We show that such an approach could be used in the planning loop to reason about unobserved causes or unlikely futures that are directly relevant to the AV's intended route.

Siddhesh Khandelwal, William Qi, Jagjeet Singh, Andrew Hartnett, Deva Ramanan• 2020

Related benchmarks

TaskDatasetResultRank
Motion forecastingArgoverse 2 Motion Forecasting Dataset (test)
Miss Rate (K=6)42
90
Trajectory PredictionnuScenes Prediction Challenge (test)
minADE (K=10)1.11
37
Trajectory PredictionArgoverse (test)
Min ADE0.8995
36
Motion forecastingArgoverse Motion Forecasting 1.1 (test)
minADE (K=1)1.82
27
Motion forecastingnuScenes official leaderboard (test)
minADE (5s)1.84
21
Trajectory PredictionArgoverse 1.0 (test)
minADE (k=6)0.9
15
Trajectory PredictioninD (val)
minADE (m)0.78
7
Trajectory PredictionDragon Lake Parking (DLP) (val)
minADE (m)0.62
7
Map-free trajectory predictionArgoverse 1.0 (val)
minADE (k=1)1.61
5
Showing 9 of 9 rows

Other info

Follow for update