TrajTok: Technical Report for 2025 Waymo Open Sim Agents Challenge

TL;DR

TrajTok is a novel trajectory tokenizer that enhances behavior generation models by combining data-driven and rule-based methods, improving realism and robustness in autonomous driving simulations, demonstrated by superior performance in the Waymo Open Sim Agents Challenge 2025.

Contribution

Introduces TrajTok, a trajectory tokenizer with a spatial-aware label smoothing technique, improving behavior prediction models for autonomous driving simulations.

Findings

Achieved a realism score of 0.7852 on the Waymo challenge.

Enhanced coverage, symmetry, and robustness in trajectory prediction.

Demonstrated superior performance over existing methods.

Abstract

In this technical report, we introduce TrajTok, a trajectory tokenizer for discrete next-token-prediction based behavior generation models, which combines data-driven and rule-based methods with better coverage, symmetry and robustness, along with a spatial-aware label smoothing method for cross-entropy loss. We adopt the tokenizer and loss for the SMART model and reach a superior performance with realism score of 0.7852 on the Waymo Open Sim Agents Challenge 2025. We will open-source the code in the future.