[Paper Review] Sampling-Based Motion Planning: A Comparative Review

Andreas Orthey, Constantinos Chamzas+1 more|arXiv (Cornell University)|Sep 22, 2023

Robot Manipulation and Learning4 citations

TL;DR

This paper presents a comprehensive comparative review of sampling-based motion planning (SBMP) algorithms, offering a historical overview, categorization of planners, and a large-scale evaluation across 24 benchmark problems. It demonstrates that no single planner excels across all scenarios, with FMT and RRT-Connect showing strong performance in most cases, while dynamic and constrained problems require specialized planners like Kinodynamic RRT or constraint-aware extensions.

ABSTRACT

Sampling-based motion planning is one of the fundamental paradigms to generate robot motions, and a cornerstone of robotics research. This comparative review provides an up-to-date guideline and reference manual for the use of sampling-based motion planning algorithms. This includes a history of motion planning, an overview about the most successful planners, and a discussion on their properties. It is also shown how planners can handle special cases and how extensions of motion planning can be accommodated. To put sampling-based motion planning into a larger context, a discussion of alternative motion generation frameworks is presented which highlights their respective differences to sampling-based motion planning. Finally, a set of sampling-based motion planners are compared on 24 challenging planning problems. This evaluation gives insights into which planners perform well in which situations and where future research would be required. This comparative review thereby provides not only a useful reference manual for researchers in the field, but also a guideline for practitioners to make informed algorithmic decisions.

Motivation & Objective

• To provide a current, comprehensive reference guide for sampling-based motion planning (SBMP) algorithms, addressing the lack of up-to-date resources for researchers and practitioners.
• To trace the historical development of SBMP from 1979 to 2023, identifying key milestones and paradigm shifts across four distinct eras: pre-sampling, sampling-advent, sampling-consolidation, and optimality/learning.
• To categorize and analyze major SBMP algorithm classes—such as PRM, RRT, BIT*, RRT*, FMT, and EST—highlighting their strengths, limitations, and suitability for different problem types.
• To compare SBMP with alternative motion generation frameworks (e.g., motion optimization, control-based planning, search-based methods), clarifying distinctions and trade-offs in completeness, optimality, and computational efficiency.
• To deliver a large-scale empirical evaluation of 10 SBMP algorithms on 24 diverse planning problems, including narrow passages, dynamic systems, and geometric constraints, to inform algorithm selection in practice.

Proposed method

• The paper conducts a historical and conceptual review of SBMP, tracing the evolution from configuration space theory (1979) to modern asymptotically optimal and learning-enhanced planners.
• It introduces a systematic categorization of SBMP algorithms based on their core structures: graph-based (e.g., PRM), tree-based (e.g., RRT), and optimal or informed variants (e.g., RRT*, BIT*, FMT).
• The authors evaluate 10 SBMP algorithms—RRT-Connect, PRM, EST, BIT*, RRT*, FMT, SST*, and kinodynamic variants—on 24 benchmark problems, measuring success rate, runtime, and solution cost.
• For constrained and dynamic scenarios, the study employs a constraint-based framework to extend planners to handle contact constraints (e.g., sphere, torus) and dynamic models (e.g., Dubin’s car, UAV with thrusters).
• The evaluation uses standardized problem instances with known difficulty, including narrow passages, high-dimensional spaces, and nonholonomic dynamics, to ensure fair and reproducible comparisons.
• The study includes ablation-style analysis on hyper-parameter sensitivity, though full tuning is noted as an open challenge, with references to existing work on parameter optimization provided.

Experimental results

Research questions

• RQ1 Which sampling-based motion planning algorithms perform best across a diverse set of 24 benchmark planning problems, including narrow passages, constraints, and dynamic systems?
• RQ2 How do the performance characteristics (success rate, runtime, solution cost) of SBMP algorithms vary across different problem types, such as geometric, kinodynamic, or constrained motion planning?
• RQ3 In what scenarios do asymptotically optimal planners (e.g., FMT, BIT*) outperform non-optimal ones (e.g., RRT-Connect), and when do they fail to converge?
• RQ4 How do SBMP methods compare to alternative motion generation frameworks like motion optimization, control-based planning, or search-based methods in terms of completeness, optimality, and computational efficiency?
• RQ5 What are the limitations of current SBMP algorithms in handling complex constraints (e.g., surface motion) or dynamic systems, and which planners are most effective in such cases?

Key findings

• FMT achieved the highest success rate in five out of six dynamic and constrained scenarios, demonstrating strong performance in complex, nonholonomic environments.
• RRT-Connect achieved 100% success in 12 out of 24 benchmark problems, including several with narrow passages, and was among the fastest in terms of runtime.
• No single planner solved all 24 scenarios; for example, the twister problem remained unsolved by all planners, highlighting persistent challenges in high-dimensional, narrow-path planning.
• In constrained motion problems (e.g., sphere, torus), EST, BIT*, and RRT-Connect achieved 100% success, while FMT found low-cost solutions in three of the six constrained cases.
• For dynamic systems such as Dubin’s car and single-thruster UAVs, only Kinodynamic RRT successfully found solutions, while optimal planners like SST* failed entirely, indicating a performance gap in dynamic settings.
• The study confirms that hyper-parameter tuning significantly affects performance, though it was not fully explored due to space constraints, and recommends further research in this area.