Memory-driven deep-reinforcement learning for autonomous robot navigation in partially observable environments

Engineering Science and Technology an International Journal, Journal Year: 2025, Volume and Issue: 62, P. 101942 - 101942

Published: Jan. 24, 2025

Language: Английский

---

Human and environmental feature-driven neural network for path-constrained robot navigation using deep reinforcement learning

Engineering Science and Technology an International Journal, Journal Year: 2025, Volume and Issue: 64, P. 101993 - 101993

Published: Feb. 25, 2025

Language: Английский

---

Informed sampling space driven robot informative path planning

Robotics and Autonomous Systems, Journal Year: 2024, Volume and Issue: 175, P. 104656 - 104656

Published: Feb. 3, 2024

Language: Английский

---

Integrating Radar-Based Obstacle Detection with Deep Reinforcement Learning for Robust Autonomous Navigation

Applied Sciences, Journal Year: 2024, Volume and Issue: 15(1), P. 295 - 295

Published: Dec. 31, 2024

This study presents an approach to autonomous navigation for wheeled robots, combining radar-based dynamic obstacle detection with a BiGRU-based deep reinforcement learning (DRL) framework. Using filtering and tracking algorithms, the proposed system leverages radar sensors cluster object points track obstacles, enhancing precision by reducing noise fluctuations. A BiGRU-enabled DRL model is introduced, allowing robot process sequential environmental data make informed decisions in unpredictable environments, achieving collision-free paths reaching goal. Simulation experimental results validate method’s efficiency adaptability, highlighting its potential real-world applications scenarios.

Language: Английский

---

Autonomous navigation and visual navigation in robot mission execution

Image and Vision Computing, Journal Year: 2025, Volume and Issue: unknown, P. 105516 - 105516

Published: March 1, 2025

Language: Английский

---

Reinforcement Learning Algorithm for Two-Leg Robot with DDPG and TD3

Lecture notes in networks and systems, Journal Year: 2025, Volume and Issue: unknown, P. 10 - 23

Published: Jan. 1, 2025

Language: Английский

---

Transformable Gaussian Reward Function for Socially Aware Navigation Using Deep Reinforcement Learning

Sensors, Journal Year: 2024, Volume and Issue: 24(14), P. 4540 - 4540

Published: July 13, 2024

Robot navigation has transitioned from avoiding static obstacles to adopting socially aware strategies for coexisting with humans. Consequently, in dynamic, human-centric environments gained prominence the field of robotics. One methods navigation, reinforcement learning technique, fostered its advancement. However, defining appropriate reward functions, particularly congested environments, holds a significant challenge. These crucial guiding robot actions, necessitate intricate human-crafted design due their complex nature and inability be set automatically. The multitude manually designed functions contains issues such as hyperparameter redundancy, imbalance, inadequate representation unique object characteristics. To address these challenges, we introduce transformable Gaussian function (TGRF). TGRF possesses two main features. First, it reduces burden tuning by utilizing small number hyperparameters that independently. Second, enables application various through transformability. exhibits high performance accelerated rates within deep (DRL) framework. We also validated simulations experiments.

Language: Английский

---

Enhancing Autonomous Robot Navigation Based on Deep Reinforcement Learning: Comparative Analysis of Reward Functions in Diverse Environments

Published: Oct. 17, 2023

Autonomous robot navigation in complex environments presents a significant challenge due to efficient decision-making for reaching goals and avoiding obstacles. This paper addresses this issue through the use of deep reinforcement learning techniques comprehensive analysis reward functions their impact on autonomous navigation. The study emphasizes importance selecting most effective achieve maximum performance variety scenarios. Moreover, we propose new mechanism that enables avoid collisions when objects move faster than robot, resulting halting its motion allow object pass before resuming course. effectiveness these is validated simulations, providing valuable insights into robustness Further details simulations can be found following link: https://youtu.be/pPQDc25vj1U

Language: Английский

---

Privacy-preserving bipartite output consensus for continuous-time heterogeneous multi-agent systems via event-triggered impulsive control

Complex & Intelligent Systems, Journal Year: 2024, Volume and Issue: 10(4), P. 5127 - 5137

Published: April 17, 2024

Abstract This paper studies the problem of differentially private bipartite output consensus in continuous-time heterogeneous multi-agent systems (MASs) characterized by antagonistic interactions. A novel hybrid privacy-preserving event-triggered impulsive protocol is introduced to protect agent’s initial information from disclosure, which involves a discrete-time transmission based on an event-triggering mechanism. Using stochastic Lyapunov method, sufficient conditions have been obtained achieve mean square with guaranteed level privacy. Furthermore, differential privacy competitive agent pairs exclusively secured proposed control scheme injecting Laplace noise. The also effectively prevents Zeno behavior imposing lower bound for intervals under all conditions. simulation example provided validate effectiveness theoretical result.

Language: Английский

---

Stranger Danger! Identifying and Avoiding Unpredictable Pedestrians in RL-based Social Robot Navigation

Published: May 13, 2024

Reinforcement learning (RL) methods for social robot navigation show great success navigating robots through large crowds of people, but the performance these learningbased tends to degrade in particularly challenging or unfamiliar situations due models' dependency on representative training data.To ensure human safety and comfort, it is critical that algorithms handle uncommon cases appropriately, low frequency wide diversity such present a significant challenge data-driven methods.To overcome this challenge, we propose modifications process encourage RL policies maintain additional caution situations.Specifically, improve Socially Attentive Learning (SARL) policy by (1) modifying systematically introduce deviations into pedestrian model, (2) updating value network estimate utilize pedestrian-unpredictability features, (3) implementing reward function learn an effective response unpredictability.Compared original SARL policy, our modified maintains similar times path lengths, while reducing number collisions 82% proportion time spent pedestrians' personal space up 19 percentage points most difficult cases.We also describe how apply other demonstrate some key high-level behaviors approach transfer physical robot.

Language: Английский

---