Learned trajectory prediction keeps an edge at low update rates

Learned versus Kinematic Trajectory Prediction under Reduced Observation Frequency

Machine LearningRobotics

Summary

Automated driving systems predict where vehicles will move based on recent position data. This study looked at how well a learned prediction model works when it gets position updates less often, like every second instead of ten times per second. The authors found that the learned model still predicts better than simple motion models, even at lower update frequencies, especially for vehicles moving straight or turning. However, for pedestrians, simple models often did better. Using training for each data frequency worked better than just tweaking a high-frequency model.

What this means in practice

  • For automated vehicle developers: Design better vehicle motion prediction systems that remain accurate even with low-frequency position updates from V2X communication.
  • For traffic simulation teams: Improve simulation accuracy by selecting prediction models according to data update rates, especially when modeling vehicle trajectories over several seconds.

Tested on one dataset.

Authors

Nukhbala Gadzhiev, Alexander Morochko

Abstract

Trajectory prediction models for automated driving are usually trained on data with a fixed sampling frequency, while positions received via vehicle-to-everything (V2X) communication arrive at 1-10 Hz. We study whether a learned predictor that uses only the observation history keeps its advantage over constant velocity (CV) and constant acceleration (CA) models when the history frequency drops. On the Argoverse 2 motion forecasting dataset, the 5 s history of the focal agent was downsampled from 10 Hz to 5, 2, and 1 Hz, and the 6 s future was predicted without maps or neighboring agents. An LSTM-based recurrent neural network (RNN) trained for each frequency was compared with the same network trained at 10 Hz and applied to sparse histories after linear interpolation. Hypotheses and thresholds were fixed before training, the main RNNs were trained with ten seeds each, and results on 24,984 test scenarios have bootstrap confidence intervals over scenarios and seeds. The average displacement error of the frequency-specific RNN grew by 3.1% at 5 Hz, 10.1% at 2 Hz, and 20.3% at 1 Hz. Contrary to our hypothesis, its advantage over the better kinematic model did not vanish on straight motion and was 24-30% on both straight motion and turns at every frequency, while for pedestrians CV was more accurate. The 10 Hz model with interpolation was at most 1.3% less accurate than the frequency-specific model down to 2 Hz but 11.5% less accurate at 1 Hz. Each hypothesis was only partly supported.