A Deep Reinforcement Learning Approach for Active SLAM

Placed, Julio A.; Castellanos, José A.

doi:10.3390/app10238386

Cited by 30 publications

(26 citation statements)

References 49 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, note that decisions will be optimal only locally and a short decision-making horizon may induce wrong behaviors [27], [114]. This strategy is typical in deep reinforcement learning (DRL) approaches [59], [115], for which local optimality is alleviated by network memorization. Following the idea that evaluating larger neighborhoods would lead to more robust decisions, in [41] authors use RRT-based paths to several configurations over the free space as the action set; and in [44] the entire environment is considered under the umbrella of continuous-domain optimization.…”

Section: B Detecting Goal Locationsmentioning

confidence: 99%

“…Optimality criteria were first used in active SLAM by Feder et al [27], where utility was computed as the area of the covariance ellipses describing the uncertainty in the robot's and map's posteriors. Since then, many active SLAM methods based on TOED have been proposed, mostly based on T -opt [35], [40] and, recently, on D-opt [59], [60]. Even so, TOED criteria are not as popular as IT approaches.…”

Section: Theory Of Optimal Experimental Design (Toed)mentioning

confidence: 99%

“…Convergence speed towards an optimal policy notably improves using parallel simulation environments, e.g., with A3C [189] and asynchronous DDPG [190], at the cost, however, of an increased training complexity. In any case, the above works use purely extrinsic reward functions, which has proved not to be solving the active SLAM problem but the obstacle avoidance one [59]. As a response, the notions of motivation and curiosity [191] were exploited to design intrinsic rewards, giving origin to curiosity-driven methods that motivate agents to visit unknown configurations [192].…”

Section: B On the Reward Function Design And The Action Setmentioning

confidence: 99%

See 2 more Smart Citations

A Survey on Active Simultaneous Localization and Mapping: State of the Art and New Frontiers

Placed¹,

Strader²,

Carrillo³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Active Simultaneous Localization and Mapping (SLAM) is the problem of planning and controlling the motion of a robot to build the most accurate and complete model of the surrounding environment. Since the first foundational work in active perception appeared, more than three decades ago, this field has received increasing attention across different scientific communities. This has brought about many different approaches and formulations, and makes a review of the current trends necessary and extremely valuable for both new and experienced researchers. In this work, we survey the state-of-the-art in active SLAM and take an in-depth look at the open challenges that still require attention to meet the needs of modern applications. After providing a historical perspective, we present a unified problem formulation and review the classical solution scheme, which decouples the problem into three stages that identify, select, and execute potential navigation actions. We then analyze alternative approaches, including belief-space planning and modern techniques based on deep reinforcement learning, and review related work on multi-robot coordination. The manuscript concludes with a discussion of new research directions, addressing reproducible research, active spatial perception, and practical applications, among other topics.

show abstract

Section: B Detecting Goal Locationsmentioning

confidence: 99%

Section: Theory Of Optimal Experimental Design (Toed)mentioning

confidence: 99%

Section: B On the Reward Function Design And The Action Setmentioning

confidence: 99%

See 1 more Smart Citation

A Survey on Active Simultaneous Localization and Mapping: State of the Art and New Frontiers

Placed¹,

Strader²,

Carrillo³

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…Some methods [14], [15] formulate this problem as the POMDP (Partially Observable Markov Decision Process) [49]. Recently, some researchers have designed learningbased policies [17]- [24] to tackle this problem. Chaplot et al [18] propose a novel module that combines a hierarchical network design and classical path planning, which significantly improves the sample efficiency and leads to better performance.…”

Section: Related Work a Active Slammentioning

confidence: 99%

Navigating to Objects in Unseen Environments by Distance Prediction

Zhu¹,

Zhao²,

Kong³

2022

Preprint

View full text Add to dashboard Cite

Object Goal Navigation (ObjectNav) task is to navigate an agent to an object instance in unseen environments. The traditional navigation paradigm plans the shortest path on a pre-built map. Inspired by this, we propose an object goal navigation framework, which could directly perform path planning based on an estimated distance map. Specifically, our model takes a birds-eye-view semantic map as input, and estimates the distance from the map cells to the target object based on the learned prior knowledge. With the estimated distance map, the agent could explore the environment and navigate to the target objects based on either human-designed or learned navigation policy. Empirical results in visually realistic simulation environments show that the proposed method outperforms a wide range of baselines on success rate and efficiency.

show abstract

“…In (Botteghi et al, 2020) and (Placed and Castellanos, 2020), the authors propose a reinforcement learning approach for active SLAM that relies on 2D LiDAR readings and information coming from the SLAM algorithm to select the best steering command, out of a discrete action space, for a mobile robot to explore different environments. While we use similar state space definition, we employ a continuous action space for obtaining smoother trajectories.…”

Section: Reinforcement Learning-based Active Slam In Roboticsmentioning

confidence: 99%

Curiosity-Driven Reinforcement Learning Agent for Mapping Unknown Indoor Environments

Botteghi

Schulte

Sırmaçek

et al. 2021

ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

Abstract. Autonomously exploring and mapping is one of the open challenges of robotics and artificial intelligence. Especially when the environments are unknown, choosing the optimal navigation directive is not straightforward. In this paper, we propose a reinforcement learning framework for navigating, exploring, and mapping unknown environments. The reinforcement learning agent is in charge of selecting the commands for steering the mobile robot, while a SLAM algorithm estimates the robot pose and maps the environments. The agent, to select optimal actions, is trained to be curious about the world. This concept translates into the introduction of a curiosity-driven reward function that encourages the agent to steer the mobile robot towards unknown and unseen areas of the world and the map. We test our approach in explorations challenges in different indoor environments. The agent trained with the proposed reward function outperforms the agents trained with reward functions commonly used in the literature for solving such tasks.

show abstract

A Deep Reinforcement Learning Approach for Active SLAM

Cited by 30 publications

References 49 publications

A Survey on Active Simultaneous Localization and Mapping: State of the Art and New Frontiers

A Survey on Active Simultaneous Localization and Mapping: State of the Art and New Frontiers

Navigating to Objects in Unseen Environments by Distance Prediction

Curiosity-Driven Reinforcement Learning Agent for Mapping Unknown Indoor Environments

Contact Info

Product

Resources

About