Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Control-ITRA: Controlling the Behavior of a Driving Model

Vasileios Lioutas, Adam Scibior, Matthew Niedoba, Berend Zwartsenberg, Frank Wood

TL;DR

Control-ITRA extends the ITRA driving-behavior model by conditioning agent actions on explicit waypoints and target speeds, enabling controllable yet realistic multi-agent trajectories. Waypoint conditioning guides agents along a sequence of coordinates, while target-speed conditioning uses FiLM-like affine transformations to modulate representations and implicitly adjust aggressiveness, with training supported by a probabilistic conditioning scheme. Experiments on large-scale real data and unseen TorchDriveEnv locations show that Control-ITRA can satisfy conditioning while maintaining realism, reducing infractions, and yielding smoother trajectories than reinforcement-learning baselines. Overall, the approach offers a practical framework for designing diverse, safety-focused driving scenarios in simulation without sacrificing human-like behavior.

Abstract

Simulating realistic driving behavior is crucial for developing and testing autonomous systems in complex traffic environments. Equally important is the ability to control the behavior of simulated agents to tailor scenarios to specific research needs and safety considerations. This paper extends the general-purpose multi-agent driving behavior model ITRA (Scibior et al., 2021), by introducing a method called Control-ITRA to influence agent behavior through waypoint assignment and target speed modulation. By conditioning agents on these two aspects, we provide a mechanism for them to adhere to specific trajectories and indirectly adjust their aggressiveness. We compare different approaches for integrating these conditions during training and demonstrate that our method can generate controllable, infraction-free trajectories while preserving realism in both seen and unseen locations.

Control-ITRA: Controlling the Behavior of a Driving Model

TL;DR

Control-ITRA extends the ITRA driving-behavior model by conditioning agent actions on explicit waypoints and target speeds, enabling controllable yet realistic multi-agent trajectories. Waypoint conditioning guides agents along a sequence of coordinates, while target-speed conditioning uses FiLM-like affine transformations to modulate representations and implicitly adjust aggressiveness, with training supported by a probabilistic conditioning scheme. Experiments on large-scale real data and unseen TorchDriveEnv locations show that Control-ITRA can satisfy conditioning while maintaining realism, reducing infractions, and yielding smoother trajectories than reinforcement-learning baselines. Overall, the approach offers a practical framework for designing diverse, safety-focused driving scenarios in simulation without sacrificing human-like behavior.

Abstract

Simulating realistic driving behavior is crucial for developing and testing autonomous systems in complex traffic environments. Equally important is the ability to control the behavior of simulated agents to tailor scenarios to specific research needs and safety considerations. This paper extends the general-purpose multi-agent driving behavior model ITRA (Scibior et al., 2021), by introducing a method called Control-ITRA to influence agent behavior through waypoint assignment and target speed modulation. By conditioning agents on these two aspects, we provide a mechanism for them to adhere to specific trajectories and indirectly adjust their aggressiveness. We compare different approaches for integrating these conditions during training and demonstrate that our method can generate controllable, infraction-free trajectories while preserving realism in both seen and unseen locations.
Paper Structure (15 sections, 9 equations, 2 figures, 5 tables, 3 algorithms)

This paper contains 15 sections, 9 equations, 2 figures, 5 tables, 3 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Trajectory Prediction:
  4. Goal-conditioned Models:
  5. Method
  6. Background: ITRA
  7. Training with Conditions
  8. Waypoint Conditioning
  9. Sampling Training Waypoints:
  10. Target Speed Conditioning
  11. Sampling Training Target Speeds:
  12. Experiments
  13. Improving Performance By Following Ground Truth Conditions
  14. Testing Out-of-domain Performance
  15. Conclusion

Figures (2)

  • Figure 1: Example ego-centric and ego-rotated birdview representations from various locations in the training set. Waypoints are shown as brown circles.
  • Figure 2: The distribution of speed values in the collected human-traffic training dataset compared to the learned speed distribution of Control-ITRA.