Bi-objective trail-planning for a robot team orienteering in a hazardous environment

Robotic path planning has evolved from classical shortest-path algorithms like Dijkstra and A* to bio-inspired heuristics such as ant colony optimization (ACO). In hazardous environments, the challenge extends beyond mere efficiency to include safety considerations, leading to models like the Team Surviving Orienteers Problem (TSOP) and robust multi-path planning. Prior works have addressed single-objective optimization, often neglecting the inherent conflict between maximizing rewards and ensuring robot survivability. Recent advances incorporate probabilistic models for path failure, but few integrate multi-objective Pareto strategies with bio-inspired algorithms, especially in complex, real-world scenarios like cultural site monitoring or disaster response.

The core issue is designing path plans that simultaneously maximize expected task rewards and robot survivability in environments with known hazard probabilities. Existing methods struggle with conflicting objectives, often sacrificing safety for reward or vice versa. The difficulty lies in efficiently exploring the vast solution space to find diverse Pareto-optimal solutions that balance these competing goals, especially under environmental uncertainties and computational constraints. Achieving this balance is critical for autonomous operations in real-world hazardous scenarios, where safety cannot be compromised, yet task success is essential.

The paper’s key innovations include: 1) adapting ant colony algorithms for bi-objective optimization with dual pheromone matrices, 2) integrating Pareto front strategies to maintain solution diversity, 3) modeling environment hazards explicitly via survival probabilities on edges, 4) designing a path redundancy mechanism to enhance robustness, and 5) validating the approach in a complex museum scenario. These innovations address the limitations of prior static or single-objective methods, enabling dynamic, risk-aware path planning that balances reward and safety, with theoretical convergence guarantees and practical validation.

• �� Model environment as a directed graph with nodes (locations) and edges (paths) with known survival probabilities. • Define reward functions for nodes and survival probabilities for edges, computing path success via product of edge survival. • Initialize dual pheromone matrices: one for reward, one for survival, and set heuristic functions based on environment data. • Deploy multiple ants, each with a parameter λ controlling the trade-off emphasis, constructing paths probabilistically using pheromone and heuristic information. • Update pheromones based on Pareto dominance, reinforcing paths that balance reward and survival. • Use iterative process to converge toward a Pareto front of solutions, ensuring diversity and optimality. • Select final paths based on human preference or operational constraints.

The experimental setup involves a museum environment modeled as a directed graph with nodes representing exhibit rooms and edges with hazard probabilities. Baseline algorithms include single-objective ACO and genetic algorithms. Metrics evaluated include expected reward, robot survival rate, Pareto front coverage, and convergence speed. Hyperparameters such as pheromone evaporation rate, number of ants, and λ distribution are tuned. Ablation studies assess the impact of heuristics and pheromone components. Results demonstrate that the proposed bi-objective ACO achieves higher reward-survivability trade-offs, with faster convergence and greater solution diversity, validating its effectiveness in complex hazard scenarios.

The proposed method outperforms baselines by achieving a 15% higher expected reward and an 85% robot survival rate in the museum scenario. Pareto front analysis shows solutions spanning a broad range of risk-reward balances, providing decision-makers with flexible options. Ablation results confirm that dual pheromone matrices and Pareto-based updates significantly improve solution quality and diversity. The method remains stable across different hazard levels, demonstrating robustness and scalability, making it suitable for real-world hazardous environment applications.

This approach is applicable in military reconnaissance, disaster response, and cultural heritage preservation, where autonomous robots must operate safely while completing complex tasks. It requires prior knowledge of hazard probabilities and reward maps, suitable for static or semi-static environments. Future integration with real-time hazard sensing and adaptive path re-planning will extend its utility to dynamic scenarios, enabling resilient autonomous operations in unpredictable hazardous conditions.

The model assumes environment hazard data are accurately known, which may not hold in real-world settings, leading to potential suboptimal paths. Computational complexity increases with environment size and number of robots, limiting scalability for large-scale deployments. The current framework does not incorporate real-time updates or dynamic hazards, necessitating future work on online learning and adaptive algorithms to handle environmental uncertainties effectively.