IR2: Implicit Rendezvous for Robotic Exploration Teams under Sparse Intermittent Connectivity (2409.04730v2)

Abstract: Information sharing is critical in time-sensitive and realistic multi-robot exploration, especially for smaller robotic teams in large-scale environments where connectivity may be sparse and intermittent. Existing methods often overlook such communication constraints by assuming unrealistic global connectivity. Other works account for communication constraints (by maintaining close proximity or line of sight during information exchange), but are often inefficient. For instance, preplanned rendezvous approaches typically involve unnecessary detours resulting from poorly timed rendezvous, while pursuit-based approaches often result in short-sighted decisions due to their greedy nature. We present IR2, a deep reinforcement learning approach to information sharing for multi-robot exploration. Leveraging attention-based neural networks trained via reinforcement and curriculum learning, IR2 allows robots to effectively reason about the longer-term trade-offs between disconnecting for solo exploration and reconnecting for information sharing. In addition, we propose a hierarchical graph formulation to maintain a sparse yet informative graph, enabling our approach to scale to large-scale environments. We present simulation results in three large-scale Gazebo environments, which show that our approach yields 6.6-34.1% shorter exploration paths when compared to state-of-the-art baselines, and lastly deploy our learned policy on hardware. Our simulation training and testing code is available at https://ir2-explore.github.io.

• The paper presents a DRL method with attention mechanisms that balances solo exploration gains with future benefits of implicit rendezvous.
• It introduces a hierarchical graph formulation combining sparse global and dense local planning to enable scalable exploration in large environments.
• Curriculum learning-based training yields a 34.1% improvement in path efficiency and enhances mapping consistency under realistic communication constraints.

Overview of $IR^2$: Implicit Rendezvous for Robotic Exploration Teams under Sparse Intermittent Connectivity

The paper presents $IR^2$, a deep reinforcement learning (DRL) approach to enhancing the efficiency and effectiveness of multi-robot exploration in large-scale environments where connectivity is both sparse and intermittent. This research addresses existing limitations of multi-robot exploration methods that often assume continuous connectivity or rely heavily on explicit rendezvous strategies that incur additional navigational overheads.

Key Contributions and Methodology

1. Deep Reinforcement Learning with Attention Mechanisms: The core of $IR^2$ involves using attention-based neural networks trained with DRL techniques. The focus is on designing policies that consider the future impacts of current actions, balancing the trade-offs between immediate solo exploration gains and future benefits of information sharing with other robots.
2. Hierarchical Graph Formulation: A novel hierarchical graph representation is proposed, which combines sparse global graphs for long-range planning with dense local graphs around each robot. This representation helps balance exploration efficiency and computational constraints, enabling scalability to larger environments without sacrificing planning robustness.
3. Curriculum Learning for Strategy Development: The training process involves a curriculum learning approach, gradually increasing the complexity of exploration environments and communication constraints. This allows the model to incrementally develop sophisticated strategies for maintaining effective exploration coverage and consistency across different scenarios.

Experimental Evaluation

The paper reports comprehensive simulation analyses, benchmarking $IR^2$ against state-of-the-art preplanned rendezvous and pursuit-based strategies. The experiments were conducted in large-scale Gazebo environments, demonstrating the efficacy of $IR^2$:

• Enhanced Exploration Efficiency: The proposed method achieves up to 34.1% improvement in exploration path efficiency compared to baseline approaches. This highlights the effectiveness of implicit rendezvous in optimizing robot paths and reducing redundant coverage.
• Improved Synchronization among Robots: There was noticeable improvement in the consistency of mapped areas among robots, indicating better coordination and information distribution within the team.
• Adaptability to Realistic Conditions: The framework incorporates realistic elements like limited connectivity, using both proximity and signal strength measures. Such adaptability reflects promising advancements towards deploying these systems in varied real-world environments.

Theoretical Implications and Future Directions

The implications of this research span both practical and theoretical domains. Practically, improving multi-robot exploration efficiency can accelerate advancements in applications such as search and rescue, planetary exploration, and subterranean mining. Theoretically, $IR^2$ introduces a pathway to further investigate decentralized learning systems that are versatile under communication constraints, exploring how reinforcement learning schemes might accommodate varying connectivity dynamics.

Future directions may include expanding the models to accommodate more complex, 3D environments, and improving communication models to incorporate factors like latency and packet loss, which could affect rendezvous dynamics. Considering these factors will enhance the robustness of $IR^2$, potentially paving the way for its application in even more challenging terrains and operational scenarios.

Overall, $IR^2$ represents a significant step towards reconciling the exploratory potential of robotic teams with practical communication limitations, using sophisticated learning techniques and novel system architectures to optimize their joint capabilities.