Abstract:
This paper presents a path-planning algorithm with obstacle avoidance based on Reinforcement Learning that is used for a stationary three-degree-of-freedom robot. Therefo...Show MoreMetadata
Abstract:
This paper presents a path-planning algorithm with obstacle avoidance based on Reinforcement Learning that is used for a stationary three-degree-of-freedom robot. Therefore, the actor-critic algorithms Deep Deterministic Policy Gradient (DDPG) and Twin Delayed Deep Deterministic Policy Gradient (TD3) are combined with Prioritized Experience Replay (PER) and tested in simulations. Further, investigations regarding different exploration strategies and network shapes are conducted. The results show that especially the combination of TD3 with PER offers a solid approach for complex path-planning in the continuous domain of the spatial three-degree-of-freedom robot. The performance could then be boosted, additionally, by enlarging the utilised feedforward-neural networks and more sophisticated exploration strategies.
Published in: 2024 American Control Conference (ACC)
Date of Conference: 10-12 July 2024
Date Added to IEEE Xplore: 05 September 2024
ISBN Information: