Adaptive Robot Navigation Using Randomized Goal Selection with Twin Delayed Deep Deterministic Policy Gradient
Year: 2025
Type: article
Abstract: The primary challenge in robotic navigation lies in enabling robots to adapt effectively to new, unseen environments. Addressing this gap, this paper enhances the Twin Delayed Deep Deterministic Policy Gradient (TD3) model’s adaptability by introducing randomized start and goal points. This approach aims to overcome the limitations of fixed goal points used in prior research, allowing the robot to navigate more effectively through unpredictable scenarios. This proposed extension was evaluated in unseen environments to validate the enhanced adaptability and performance of the TD3 model. The experimental results highlight improved flexibility and robustness in the robot’s navigation capabilities, demonstrating the model’s ability to generalize effectively to unseen environments. In addition, this paper presents an overview of the TD3 algorithm’s architecture and principles, with an emphasis on key components like the actor and critic networks, their updates, and mitigation of overestimation bias. The goal is to provide a clearer understanding of how the TD3 framework is adapted and utilized in this study to achieve improved performance.
Source: Robotics
Authors: Romisaa Ali, Sedat Dogru, Lino Marques, Marcello Chiaberge
Institutions: Politecnico di Torino, University of Coimbra
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,172
Award: G3092181242
OpenAlex ID: W4409010269