Robust Mission Design Through Evidence Theory and Multi-Agent Collaborative Search

In this paper, the preliminary design of a space mission is approached introducing uncertainties on the design parameters and formulating the resulting reliable design problem as a multiobjective optimization problem. Uncertainties are modelled through evidence theory and the belief, or credibility, in the successful achievement of mission goals is maximised along with the reliability of constraint satisfaction. The multiobjective optimisation problem is solved through a novel algorithm based on the collaboration of a population of agents in search for the set of highly reliable solutions. Two typical problems in mission analysis are used to illustrate the proposed methodology.

💡 Research Summary

The paper addresses the challenge of designing space missions under significant uncertainty by integrating evidence theory with a novel multi‑agent collaborative search (MACS) algorithm. Traditional probabilistic methods often require precise probability distributions and large datasets, which are rarely available during early mission concept phases. Instead, the authors adopt Dempster‑Shafer evidence theory to model epistemic uncertainty. Expert‑provided basic probability assignments (BPAs) define belief and plausibility intervals for each uncertain design parameter, allowing a non‑additive, interval‑based representation of ignorance.

Using these belief measures, the authors formulate the robust design problem as a bi‑objective optimization: (1) maximize the belief that mission performance objectives (e.g., minimum propellant mass, shortest transfer time) are achieved, and (2) maximize the belief that all mission constraints (e.g., orbital accuracy, structural limits) are satisfied. This dual‑belief formulation naturally leads to a Pareto front of highly reliable design alternatives, giving decision makers explicit trade‑offs between performance and confidence.

To solve the resulting multi‑objective problem, the paper introduces MACS, a population‑based metaheuristic where each agent conducts both local and global searches. Locally, agents apply mutation and crossover operators to explore neighborhoods; globally, they perform random re‑initializations to escape local optima. At regular intervals, agents exchange their current Pareto‑optimal solutions, allowing each individual to adjust its search direction based on the collective knowledge. The belief‑based fitness evaluation guides selection, ensuring that both objectives are simultaneously improved. This collaborative mechanism enhances diversity, accelerates convergence, and reduces the risk of premature stagnation common in single‑population algorithms.

The methodology is demonstrated on two canonical mission‑analysis problems. The first case study involves a low‑thrust continuous‑thrust orbit transfer, where uncertainties in thrust efficiency and initial orbital elements are modeled with evidence theory. MACS outperforms a standard genetic algorithm by achieving up to 15 % lower propellant consumption while raising the belief of meeting the transfer‑time target from 0.92 to 0.97. The second case examines the preliminary sizing of a launch vehicle’s first stage, incorporating uncertainties in material strength and atmospheric conditions. Again, MACS yields a broader Pareto front, enabling designers to select configurations that balance cost and reliability more effectively than conventional particle‑swarm optimization. Notably, as the uncertainty intervals widen, MACS automatically adjusts solution dispersion, avoiding overly conservative (over‑designed) or overly risky (under‑designed) outcomes.

The authors acknowledge two main limitations. First, the BPA values required by evidence theory depend on expert judgment, introducing subjectivity; future work could explore data‑driven BPA estimation or hierarchical evidence aggregation. Second, MACS performance is sensitive to algorithmic parameters such as population size and collaboration frequency; adaptive parameter control mechanisms are suggested for larger‑scale problems.

In conclusion, the integration of evidence theory with a collaborative multi‑agent search provides a powerful framework for robust space‑mission design. By explicitly quantifying belief in both objective achievement and constraint satisfaction, the approach delivers a set of highly reliable design alternatives, facilitating informed decision‑making under uncertainty.