Autonomous robot navigation within a dynamic environment is a complicated issue since environmental factors keep on changing, safety remains a factor, and issues of data privacy concern are also on the increase. The existing reinforcement learning (RL) navigation systems mainly focus on path performance and avoidance of collisions but do not focus on privacy protection, adaptation learning stability, and real deployment. This research aims to overcome these constraints by suggesting a novel framework Secure and Privacy-Preserving Hyperparameter-Tuned RL Model (SPHTRLM) to the efficient generation of path plans in grid ecosystems with dynamic environments. The framework incorporates adjusted Q-learning with federated learning (FL) based distributed updates, refined differentiated privacy, minimal encrypted parameter exchange, adaptive reward shaping and automatic hyperparameter optimization. In a further attempt to enhance practicability, the proposed architecture also embraces mobility conscious aggregation and heterogeneous model support of resource-limited robotic platforms. The suggested SPHTRLM has a success rate of (95% ± 2%), and it is better than the comparable one Q-learning (87% ± 4%) and Deep RL (DRL) baselines (88%) when these methods were evaluated under the same condition. The framework minimizes distances to the average path with a reduction of 20–25% and convergence is speeded up by around 35% compared to normal Q-learning. When the obstacles are very thick then the collision rate becomes and the obstacle reduces to 0.08, and the safety of the navigation process improves. Although there are additional privatization mechanisms, the computational costs are minimal (8–12%), and the average decision time is 110–125 ms, which meets the real-time operational capabilities. Privacy analysis with formally stated membership inference and reconstruction attacks provide status of attack rate less than 5% attack success with both white and black box adversary. These findings underscore that SPHTRLM is a feasible way of achieving the goals of ensuring navigation, learning consistency, safety as well as privacy protection to give credible acceptance to using autonomous robotic systems in dynamic and data-sensitive environment.
Dewangan et al. (Thu,) studied this question.