Self-Evolving Agentic Logic Algorithms for Dynamic Multi-Goal Navigation

Abstract

Effective multi-goal navigation in autonomous systems represents one of the key challenges faced by researchers today. In such scenarios, the agents need to successfully navigate towards achieving multiple goals within their environments that have moving obstacles. Traditional methods like path-planning techniques and reinforcement learning techniques often fail when it comes to providing real-time adaptive capabilities while achieving multiple goals. In this study, we propose a novel algorithm called Self-Evolving Agentic Logic (SEAL), which uses techniques of goal prioritization, agentic reasoning, and self-evolution to allow autonomous agents to adopt effective strategies for successful multi-goal navigation. SEAL has been tested in a virtual 2D environment that has dynamic and static obstacles with random placement of multiple goals. Metrics used were Success Rate (SR), Path Efficiency (PE), Computational Cost (CC), and Goal Completion Time (GCT). The proposed algorithm was benchmarked against existing algorithms such as A*, Deep Q-Networks (DQN), and conventional Agentic Logic. The SEAL algorithm demonstrated the highest success rate (96.5%), path efficiency (92.3%), and goal completion time (18.2 s), while still ensuring relatively low computational costs (12.4 ms per decision). Comparison studies showed significant statistical improvement over all baselines (p<0.05), which indicated improved adaptability and robustness to changes in dynamic multi-goal environments. The SEAL algorithm is a powerful and flexible tool for multi-goal navigation. It can be easily scaled to larger problems due to its modularity and self-evolutionary approach. Future research may consider the application of SEAL to multi-agent scenarios, robotics, and reinforcement learning frameworks.