Abstract
Exploration of unknown environments is an important challenge in the field of robotics. While a single robot can achieve this task alone, evidence suggests it could be accomplished more efficiently by groups of robots, with advantages in terms of terrain coverage as well as robustness to failures. Exploration can be guided through belief maps, which provide probabilistic information about which part of the terrain is interesting to explore (either based on risk management or reward). This process can be centrally coordinated by building a collective belief map on a common server. However, relying on a central processing station creates a communication bottleneck and single point of failure for the system. In this paper, we present Distributed Online Risk-Aware (DORA) Explorer, an exploration system that leverages decentralized information sharing to update a common risk belief map. DORA Explorer allows a group of robots to explore an unknown environment discretized as a 2D grid with obstacles, with high coverage while minimizing exposure to risk, effectively reducing robot failures.
Introduction
In critical scenarios such as search-and-rescue or planetary exploration, multi-robot teams are deployed to map unknown environments. While these systems offer inherent resilience, standard algorithms like Frontier-Based Exploration (FBE) often prioritize coverage at the expense of safety. In hazardous environments containing radiation or toxic gases, this "coverage-first" approach can lead to rapid robot failure, ultimately jeopardizing the mission. Furthermore, centralized coordination introduces single points of failure. DORA addresses these challenges through a decentralized, risk-aware framework designed to maintain swarm population while achieving high coverage.
Methodology
DORA's architecture is built upon three core pillars that enable decentralized operation and risk mitigation.
Distributed Belief Maps (DBMs)
The system utilizes "virtual stigmergy" implemented via Conflict-free Replicated Data Types (CRDTs). Robots share environmental data—such as risk levels and last-visited timestamps for grid cells—as key-value pairs. This peer-to-peer synchronization allows each robot to maintain a local belief map that reflects the collective knowledge of the swarm without requiring a central server.
Risk and Information Modeling
Risk is modeled as a scalar field (e.g., radiation intensity), where the probability of failure increases with intensity. Information gain is modeled using an exponential distribution; unvisited or long-neglected cells offer higher utility, incentivizing exploration of "fresh" or unknown areas.
Gradient-Based Control Law
Instead of computationally expensive global path planning, DORA employs a lightweight movement vector $m_i$ for each robot $i$. This vector is a weighted sum of three local gradients:
- Risk Avoidance ($\alpha$): Directs the robot away from high-risk zones.
- Exploration ($\beta$): Pulls the robot toward areas with high information potential.
- Obstacle Avoidance ($\gamma$): Prevents physical collisions.
The gradient is calculated based on the robot's immediate 3x3 neighborhood in the discretized grid, as shown in the figure above.
Experimental Results
The DORA Explorer was evaluated through ARGoS simulations with up to 20 KheperaIV robots and physical laboratory experiments.
Survival and Coverage
In simulations, DORA demonstrated a significant advantage in robot survival. While FBE often resulted in total swarm loss in high-risk environments, DORA typically preserved 90% of the robots. Crucially, this safety did not come at the cost of long-term coverage. By keeping the workforce alive, DORA achieved terrain coverage comparable to FBE, which eventually stalls once its robots fail.
Decentralization and Scalability
The use of CRDTs ensured robustness against network disconnections. Communication overhead remained constant per robot, scaling linearly with the local neighborhood size rather than the total swarm size, making DORA suitable for large-scale deployments.
Discussion and Future Work
DORA demonstrates that risk avoidance is a performance strategy, not just a safety measure. However, the current model assumes risks are perceivable before failure. Future work will explore "failure memory"—using the location of robot deaths to infer hidden hazards—and adaptive risk-taking profiles to balance urgency and safety dynamically.
Summary updated