Deep Active Localization

Gottipati, Sai Krishna; Seo, Keehong; Bhatt, Dhaivat; Mai, Vincent; Murthy, K. S. R.; Paull, Liam

doi:10.1109/lra.2019.2932575

Cited by 31 publications

(15 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, ANL [65] bases an assumption that the transition functions in ANL are deterministic, which does not apply well to real robot. Therefore, Gottipati et al [66] proposed a hierarchical likelihood estimation approach which decouples the resolution of the likelihood resolution from the distance that the robot travels, and used advantage actor-critic (A2C) to accomplish localization task for real robot.…”

Section: A Direct Drl Vnavigationmentioning

confidence: 99%

A Survey on Visual Navigation for Artificial Agents With Deep Reinforcement Learning

Zeng

Wang

2020

IEEE Access

View full text Add to dashboard Cite

Visual navigation (vNavigation) is a key and fundamental technology for artificial agents' interaction with the environment to achieve advanced behaviors. Visual navigation for artificial agents with deep reinforcement learning (DRL) is a new research hotspot in artificial intelligence and robotics that incorporates the decision making of DRL into visual navigation. Visual navigation via DRL, an end-to-end method, directly receives the high-dimensional images and generates an optimal navigation policy. In this paper, we first present an overview on reinforcement learning (RL), deep learning (DL) and deep reinforcement learning (DRL). Then, we systematically describe five main categories of visual DRL navigation: direct DRL vNavigation, hierarchical DRL vNavigation, multi-task DRL vNavigation, memory-inference DRL vNavigation and vision-language DRL vNavigation. These visual DRL navigation algorithms are reviewed in detail. Finally, we discuss the challenges and some possible opportunities to visual DRL navigation for artificial agents.

show abstract

Section: A Direct Drl Vnavigationmentioning

confidence: 99%

A Survey on Visual Navigation for Artificial Agents With Deep Reinforcement Learning

Zeng

Wang

2020

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Tai and Liu [19] Navigation DQN Extrinsic Mirowski et al [20] Navigation A3C Extrinsic Wen et al [22] Navigation D3QN Extrinsic Zhelo et al [26] Goal navigation A3C Intrinsic Oh and Cavallaro [27] Goal navigation A3C Intrinsic Tai et al [35] Goal navigation ADDPG Intrinsic Zhu et al [33] Frontier selection A3C Intrinsic Niroui et al [34] Frontier selection A3C Intrinsic Gottipati et al [31] Active localization A2C Posterior belief Chaplot et al [30] Active localization A3C Posterior belief Chen et al [29] Exploration DQN Entropy Chen et al [32] Active…”

Section: Partial Observabilitymentioning

confidence: 99%

“…In [28,29] actions are selected so as to maximize the entropy reduction (information gain or Kullback-Leibler divergence) of a 2-dimensional occupancy grid map; either by learning in a supervised fashion with labelled cells, or by introducing a metric of it in the DRL's reward function, respectively. Another approach to solve the active localization problem that also acts directly on the reward function design was proposed in [30,31], where the entropy reduction was achieved by including on it the maximum likelihood (ML) of being in any state (belief accuracy). Recently, Chen et al [32] trained both DQN and A2C agents using the underlying SLAM pose-graph through Graph Neural Networks (GNN).…”

Section: Introductionmentioning

confidence: 99%

A Deep Reinforcement Learning Approach for Active SLAM

Placed

Castellanos

2020

Applied Sciences

View full text Add to dashboard Cite

In this paper, we formulate the active SLAM paradigm in terms of model-free Deep Reinforcement Learning, embedding the traditional utility functions based on the Theory of Optimal Experimental Design in rewards, and therefore relaxing the intensive computations of classical approaches. We validate such formulation in a complex simulation environment, using a state-of-the-art deep Q-learning architecture with laser measurements as network inputs. Trained agents become capable not only to learn a policy to navigate and explore in the absence of an environment model but also to transfer their knowledge to previously unseen maps, which is a key requirement in robotic exploration.

show abstract

“…Such approaches can be computationally intensive in general environments. In a novel and promising approach, the navigation policy for active localization is learned using reinforcement learning [Gottipati et al 2019]. Note that active localization is related to the more general problem of simultaneous location and mapping (SLAM) [Rekleitis et al 2006].…”

Section: Related Workmentioning

confidence: 99%

Improving Monte Carlo Localization with Strategic Navigation Policies and Optimal Landmark Placement

Júnior

Figueiredo

2020

REIC

View full text Add to dashboard Cite

An important problem in robotics is to determine and maintain the position of a robot that moves through a known environment with indistinguishable landmarks. This problem is made difficult due to the inherent noise in robot movement and sensor readings. Monte Carlo Localization (MCL) is a frequently used technique to solve this problem, and its performance intuitively depends on how the robot explores the environment and the position of the landmarks. In this paper, we propose a navigation policy to reduce the number of steps required by the robot to find its location together with the optimal landmark placement for this policy. This proposal is evaluated and compared against other policies using two specific metrics that indicate its superiority.

show abstract

Deep Active Localization

Cited by 31 publications

References 24 publications

A Survey on Visual Navigation for Artificial Agents With Deep Reinforcement Learning

A Survey on Visual Navigation for Artificial Agents With Deep Reinforcement Learning

A Deep Reinforcement Learning Approach for Active SLAM

Improving Monte Carlo Localization with Strategic Navigation Policies and Optimal Landmark Placement

Contact Info

Product

Resources

About