Dynamic Graph-Based Models in Robotics: Addressing Real-World Challenges
This review explores how dynamic environments impact graph-based world models in robotics.
Graph-based models have become essential in robotics for representing internal world knowledge, particularly through factor and scene graphs. While many existing models focus on static environments, this review highlights the need to adapt these models to dynamic real-world conditions.
Key Aspects of Dynamic Graph Models
The review is structured around three main themes:
- Suitable representations for dynamic environments
- Pipelines for constructing and updating these representations
- Utilization of these models in practical applications
It emphasizes hybrid models that integrate both factor and scene graphs to better capture dynamics.
Challenges and Trends
Emerging trends and challenges include:
- Uncertainty propagation from perception to representation
- Observability of dynamic entities
- Scalable lifelong maintenance of models
- Need for datasets and evaluation protocols