Perfect Prediction or Plenty of Proposals? What Matters Most in Planning for Autonomous Driving
Aron Distelzweig, Faris Janjo�, Oliver Scheel, Sirish Reddy Varra, Raghu Rajan, Joschka Boedecker
AI summary
Problem
It remains unclear whether Integrated Prediction and Planning (IPP) methods actually benefit from accurate predictions or effectively utilize future behavior information, especially in complex, interactive driving scenarios.
Approach
The authors evaluate state-of-the-art IPP methods under perfect, learned, and no-prediction conditions, then introduce a rule-based proposal generation method that prioritizes diverse, realistic trajectory candidates over prediction accuracy.
Key results
- Perfect predictions fail to improve planning performance in existing IPP methods.
- Imitation learning planners struggle to generate realistic proposals, often underperforming simple baselines.
- A novel rule-based proposal generator produces more diverse and feasible driving maneuvers.
- The proposed method sets new state-of-the-art results on interactive and out-of-distribution benchmarks.
Why it matters
It redirects autonomous driving research away from over-reliance on prediction accuracy and toward robust proposal generation, critical for safe navigation in complex, interactive traffic.
Abstract
Traditionally, prediction and planning in au- tonomous driving (AD) have been treated as separate, sequen- tial modules. Recently, there has been a growing shift towards tighter integration of these components, known as Integrated Prediction and Planning (IPP), with the aim of enabling more informed and adaptive decision-making. However, it remains unclear to what extent this integration actually improves planning performance. In this work, we investigate the role of prediction in IPP approaches, drawing on the widely adopted Val14 benchmark, which encompasses more common driving scenarios with relatively low interaction complexity, and the interPlan benchmark, which includes highly interactive and out-of-distribution driving situations. Our analysis reveals that even access to perfect future predictions does not lead to better planning outcomes, indicating that current IPP methods often fail to fully exploit future behavior information. Instead, we focus on high-quality proposal generation, while using predictions primarily for collision checks. We find that many imitation learning-based planners struggle to generate realistic and plausible proposals, performing worse than PDM—a simple lane-following approach. Motivated by this observation, we build on PDM with an enhanced proposal generation method, shifting the emphasis towards producing diverse but realistic and high-quality proposals. This proposal-centric approach significantly outperforms existing methods, especially in out-of- distribution and highly interactive settings, where it sets new state-of-the-art results.