GRACE: A Unified 2D Multi-Robot Path Planning Simulator & Benchmark for Grid, Roadmap, And Continuous Environments
New benchmark enables direct comparison of planning algorithms across three abstraction levels on identical tasks.
A research team including Chuanlong Zang, Anna Mannucci, and four others has introduced GRACE, a novel simulator and benchmark designed to solve a critical problem in robotics research: the inability to fairly compare different Multi-Robot Path Planning (MRPP) algorithms. Existing tools force researchers to choose between high-fidelity continuous simulators that are computationally heavy and difficult to compare, or simplified grid-based models that scale well but lack realism. GRACE breaks this deadlock by instantiating the exact same planning task—like navigating a warehouse or factory floor—across three distinct environmental representations: discrete grids, topological roadmaps, and continuous 2D space.
This unified framework enables, for the first time, commensurate "apples-to-apples" comparisons. Researchers can now run an algorithm designed for grid worlds and another designed for continuous motion on the identical problem instance within GRACE, using its explicit, reproducible transformation operators. The team's initial empirical results, using public maps and representative planners, already quantify a key trade-off: continuous-fidelity MRMP solvers achieve more realistic paths but are slower, while grid and roadmap planners sacrifice some accuracy for far greater scalability. By consolidating representation, execution, and evaluation into one platform, GRACE aims to make cross-representation studies transparent and reproducible.
The simulator's standardized evaluation protocol measures standard metrics like makespan (total time to complete all paths) and success rate across all abstraction levels. This methodological rigor is crucial for advancing the field beyond isolated benchmarks and toward research that genuinely translates to real-world robotics deployment. The code is slated for release soon, and the paper is accepted for presentation at the prestigious ICRA 2026 conference, signaling its anticipated impact on the robotics community.
- Enables direct comparison of planning algorithms across grid, roadmap, and continuous 2D environment models on identical tasks.
- Quantifies the fidelity-speed trade-off, showing continuous planners are more accurate but slower, while grid planners scale further.
- Provides a unified evaluation protocol to advance reproducible research in Multi-Agent Pathfinding (MAPF) and Multi-Robot Motion Planning (MRMP).
Why It Matters
Provides a standardized testbed to accelerate robotics research, enabling fair algorithm comparisons and smoother translation from simulation to real-world deployment.