Massive Parallel Deep Reinforcement Learning for Active SLAM

Martín Arce Llobera; Julio A. Placed; Mariano De Paula; Pablo De Cristóforis

cs.ROMar 26, 2026

Massive Parallel Deep Reinforcement Learning for Active SLAM

Authors: Martín Arce Llobera, Julio A. Placed, Mariano De Paula, Pablo De Cristóforis

Organizations: Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires, Ciudad Autonoma de Buenos Aires, C1428EGA, Argentina · Instituto Tecnológico de Aragón (ITA) and the University of Zaragoza, María de Luna 3-7, Zaragoza, Spain

Abstract

Recent advances in parallel computing and GPU acceleration have created new opportunities for computation-intensive learning problems such as Active SLAM -- where actions are selected to reduce uncertainty and improve joint mapping and localization. However, existing DRL-based approaches remain constrained by the lack of scalable parallel training. In this work, we address this challenge by proposing a scalable end-to-end DRL framework for Active SLAM that enables massively parallel training. Compared with the state of the art, our method significantly reduces training time, supports continuous action spaces and facilitates the exploration of more realistic scenarios. It is released as an open-source framework to promote reproducibility and community adoption.

Figures & tables

Fig. 1 : Parallel DRL training with 750 robots (cyan). The proposed pipeline leverages NVIDIA Isaac Sim/Lab for Active SLAM training in a GPU-accelerated environment.

Method	Task	Training Architecture	Reward Design	Partial Obs.	Generalization Unknown Env.	Continuous Action Space	Realistic Env.	Parallel Training
Niroui et al. (2019) [ 10 ]	Frontier selection	A3C	Intrinsic	✓	✓	✗	✓	✗
Li et al. (2019) [ 11 ]	Frontier selection	DQN	Map entropy	✓	✓	✗	✗	✗
Chen et al. (2020) [ 9 ]	Frontier selection	DQN	T-opt	✓	✓	✗	✗	✗
Cimurs et al. (2021) [ 12 ]	Local Motion	TD3	Intrinsic	✓	✓	✓	✓	✓
Chaplot et al. (2020) [ 13 ]	Active localization	A3C	Posterior belief	✗	✓	✗	✓	✗
Placed et al. (2020) [ 5 ]	Active SLAM	D3QN	D-opt	✓	✓	✗	✗	✗

TABLE I : Comparison of the proposed approach and existing methods organized by the task performed by the DRL module.

Fig. 2 : Overview of the proposed uncertainty-aware Active SLAM framework.

Panel of Figure 4 — Fig. 3 : Isaac Sim/Lab testing environments. Env. 2 and Env. 3 were recreated following [ 5 ] . Env. 4 was taken from [ 21 ] .

Panel of Figure 5 — Fig. 4 : Training curves of our pipeline using $750$ parallel agents in environment 1 . This figure shows that tens of millions of ASLAM training timesteps can be achieved in a few hours with our pipeline.

Fig. 5 : Correlation between RMSE (estimated pose - GT) and $U_{t}$ . This graph was computed in Environment $1$ .

Panel of Figure 7 — Fig. 6 : Occupancy grids created by our fine-tuned uncertainty-aware agent using slam_toolbox and rviz2 . Red circles indicate the starting position, whereas red arrows indicate the trajectory followed by the robot.

Panel of Table 8 — TABLE II : Test results across three environments letting the agent move for a maximum of $3000$ timesteps ( $5$ minutes) or until a collision occurs. The results were averaged across $100$ runs. The collision rate represents actual physical collisions against obstacles.

RL for Robotics

Reinforcement learning for robotics trains robots to act through interaction, from locomotion and navigation to dexterous manipulation.

See all 252 papers

A Survey on Reinforcement Learning Applications in SLAM
Aug 26, 2024Mohammad Dehghani Tezerjani, Mohammad Khoshnazar, Mohammadhamed Tangestanizadeh +2Simultaneous Localization and Mapping RL for Robotics
Computer Science and Engineering, University of North Texas, Denton, USA · Institute of Artificial Intelligence, University of Bremen, Bremen, Germany · Computer Science and Engineering, University of California Santa Cruz, Santa Cruz, USA +1
Dream-SLAM: Dreaming the Unseen for Active SLAM in Dynamic Environments
Feb 25, 2026Xiangqi Meng, Pengxu Hou, Zhenjun Zhao +4Simultaneous Localization and Mapping Visual SLAM
Thrust of Robotics and Autonomous Systems, Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China. · University of Zaragoza, Zaragoza, Spain. · School of Computation, Information and Technology, Technical University of Munich, Munich, Germany. +1
AMB3R-SLAM: Kilometer-scale SLAM with Hierarchical Backend
Sep 17, 2026Hengyi Wang, Lourdes AgapitoVisual SLAM
Department of Computer Science, University College London

Aug 26, 2024J/K move · Enter open · S save

A Survey on Reinforcement Learning Applications in SLAM

Mohammad Dehghani Tezerjani, Mohammad Khoshnazar, Mohammadhamed Tangestanizadeh, Arman Kiani+1

Computer Science and Engineering, University of North Texas, Denton, USA · Institute of Artificial Intelligence, University of Bremen, Bremen, Germany · Computer Science and Engineering, University of California Santa Cruz, Santa Cruz, USA+1

Simultaneous Localization and Mapping RL for Robotics Robotic RL Loop Closure Detection

Open paper AskPDF arXiv

Simultaneous localization and mapping (SLAM) allows a mobile robot or autonomous vehicle to build a map of an unknown environment while estimating its own pose within that map. Reinforcement learning (RL), in which an agent learns a decision policy from interaction and reward, has been applied to decide how such systems move, explore, and recognize places they have visited before. This survey reviews the applications of RL in SLAM. We first distinguish passive SLAM, in which the robot's motion is not chosen by the SLAM system, from active SLAM, in which it is, and summarize the sensors that provide the input to SLAM. We then introduce the RL methods used in this literature, from value-based and policy-based methods to actor-critic and deep RL. Next, we classify RL applications in SLAM into three categories: path planning, including environment exploration and obstacle avoidance; loop closure detection; and active SLAM. Thirteen representative studies are compared in terms of their simulation environment, deep learning method, SLAM method, and RL algorithm. Most of these studies are evaluated mainly in simulation, and value-based methods from the deep Q-network family are the most common. Finally, we discuss the challenges of applying RL to SLAM, namely computational demands, safety, generalization, high-dimensional state and action spaces, sample efficiency, and sensor and actuator delays, and we outline directions for future research.

Abstract

RL for Robotics

Explore similar work

A Survey on Reinforcement Learning Applications in SLAM

Dream-SLAM: Dreaming the Unseen for Active SLAM in Dynamic Environments

AMB3R-SLAM: Kilometer-scale SLAM with Hierarchical Backend