Diffusion Model for Planning: A Systematic Literature Review

Toshihide Ubukata,Jialong Li,Kenji Tei

2024-08-16

Abstract:Diffusion models, which leverage stochastic processes to capture complex data distributions effectively, have shown their performance as generative models, achieving notable success in image-related tasks through iterative denoising processes. Recently, diffusion models have been further applied and show their strong abilities in planning tasks, leading to a significant growth in related publications since 2023. To help researchers better understand the field and promote the development of the field, we conduct a systematic literature review of recent advancements in the application of diffusion models for planning. Specifically, this paper categorizes and discusses the current literature from the following perspectives: (i) relevant datasets and benchmarks used for evaluating diffusion modelbased planning; (ii) fundamental studies that address aspects such as sampling efficiency; (iii) skill-centric and condition-guided planning for enhancing adaptability; (iv) safety and uncertainty managing mechanism for enhancing safety and robustness; and (v) domain-specific application such as autonomous driving. Finally, given the above literature review, we further discuss the challenges and future directions in this field.

Machine Learning,Artificial Intelligence,Robotics

What problem does this paper attempt to address?

The paper aims to address the following issues:

1. **Application of Diffusion Models in Planning Tasks**: The paper conducts a comprehensive examination of the latest advancements in the field of planning using diffusion models through a Systematic Literature Review (SLR). Specifically, it discusses the growth of related research since 2023. The paper categorizes and discusses the current literature, including:
- Relevant datasets and benchmarks (such as D4RL, Gym-MuJoCo, etc.) used to evaluate planning methods based on diffusion models.
- Fundamental research (e.g., improvements in sampling efficiency).
- Skill-centric and condition-guided planning methods to enhance adaptability.
- Safety and uncertainty management mechanisms to improve safety and robustness.
- Applications in specific domains, such as autonomous driving.

2. **Challenges Brought by Technological Diversity and Broad Applications**: Due to the technological diversity and wide range of application scenarios of diffusion models, comprehensively understanding planning methods based on diffusion models is a challenge. The paper aims to systematically organize existing research to provide researchers and practitioners with a comprehensive overview, making it easier for them to understand the latest advancements in the field of planning using diffusion models.

3. **Future Research Directions and Challenges**: Based on the aforementioned literature review, the paper also discusses future challenges in the field and proposes possible research directions to further advance the development of this area.