CORAL: Curriculum-Optimized Reward Adaptation for LiDAR-Based Goal-Directed Urban Driving

CORAL: Curriculum-Optimized Reward Adaptation for LiDAR-Based Goal-Directed Urban Driving

|Authors: Anisa Saleem, Duksu Kim

|Venue: arXiv preprint | Year: 2026

|arXiv: 2608.14332v1

Abstract

Reinforcement learning is promising for autonomous urban driving, but long-horizon goal-directed navigation asks a policy to acquire several competing behaviors at once—reaching a distant goal, tracking a route, avoiding obstacles, obeying signals—and a fixed objective gives no order in which to learn them. This paper presents CORAL, which advances two schedules together: a five-stage curriculum that progressively lengthens routes and tightens behavioral constraints, and a stage-aware reward whose component weights shift emphasis from mission progress toward route following, safety, smoothness, and rule compliance as the task hardens. The policy is a multi-stream actor-critic network trained with Proximal Policy Optimization (PPO) in CARLA on a compact 99-dimensional state pairing a polar LiDAR histogram with vehicle telemetry, ego-frame route geometry, and traffic-rule indicators—no point-cloud encoder, no bird’s-eye-view rasterization. Against two PPO baselines under an identical protocol, CORAL reaches the goal in all twenty evaluation episodes on the longest routes under the full set of behavioral constraints, where the baselines reach 5% and 10%; a factorial ablation shows that neither schedule alone matches their combination: removing either lowers both success and route completion, and disabling both drops success to 55%. Trained in one town, the policy transfers zero-shot to seven unseen towns, succeeding in 68-98% of episodes on routes of the same 100-150 m length, with mean lateral deviation below 0.35 m.

Key Contributions

• Reinforcement learning is promising for autonomous urban driving, but long-horizon goal-directed …
• This paper presents CORAL, which advances two schedules together: a five-stage curriculum that pr…
• The policy is a multi-stream actor-critic network trained with Proximal Policy Optimization (PPO)…
• Against two PPO baselines under an identical protocol, CORAL reaches the goal in all twenty evalu…
• Trained in one town, the policy transfers zero-shot to seven unseen towns, succeeding in 68-98% o…

Topics

• reinforcement-learning
• planning
• benchmark

Code & Data

No code repository linked in paper metadata.

BibTeX

@article{Saleem2026_260814332v1,
  title     = {CORAL: Curriculum-Optimized Reward Adaptation for LiDAR-Based Goal-Directed Urban Driving},
  author    = {Anisa Saleem and Duksu Kim},
  year      = {2026},
  eprint    = {2608.14332v1},
  archivePrefix = {arXiv},
  primaryClass  = {cs.RO},
  url       = {https://arxiv.org/abs/2608.14332v1}
}