Study Compares Eight Self-Driving Motion Planners Using Unified CARLA Test Platform

A new comparative analysis brings together eight representative motion-planning approaches from the major autonomous driving research ecosystems and evaluates them side-by-side on a common testing ground.

The Challenge of Fair Comparison#

The autonomous driving field has developed multiple competitive benchmarks—CARLA, nuPlan, and the Waymo Open Dataset chief among them—each with its own datasets and evaluation methods. This fragmentation makes it difficult to directly compare different planning algorithms, since each operates under different conditions and metrics. A unified evaluation framework could help researchers understand which techniques work best and why.

Testing Eight Leading Methods#

To create a level playing field, researchers adopted CARLA Leaderboard v2.1 as the standard evaluation platform and assessed eight prominent planning methods: TF++, InterFuser, TCP, PDM-Lite, MTR+MPC, CaRL, PlanT 2.0, and Diffusion planner. These were drawn from the major benchmark ecosystems to represent the current state of the field.

What the Study Reveals#

By running all eight approaches through the same test scenarios in CARLA, the team pinpointed shared patterns—including recurring strengths, common failure modes, and gaps in existing solutions. The analysis draws connections between methods developed in isolation, offering insights into which strategies prove most reliable across diverse driving situations and environmental conditions.

Looking Ahead#

The comparative case study identifies directions for future research by clarifying where current motion-planning techniques excel and where they fall short. The findings suggest priorities for the next generation of autonomous driving algorithms.