Active Simultaneous Localization and Mapping in Uncertain Environments

Active simultaneous localisation and mapping (SLAM) refers to the integration of autonomous decision-making with the simultaneous construction of a map and the estimation of a mobile agent’s pose in environments where prior information is scarce or unreliable. Unlike passive SLAM, where data collection follows a predetermined path, active SLAM selects trajectories that maximise the information gathered, thereby reducing uncertainty in both the map and pose estimates. Central to this approach are strategies drawn from information theory, optimal experimental design and decision-theoretic planning. Robots evaluate candidate actions according to metrics such as expected information gain or reduction in entropy, often framing the task as a form of planning under uncertainty within a partially observable Markov decision process. Techniques range from frontier-based exploration and pose-graph optimisation to belief-space planning and learning-based controllers. Advances in sensor technology, computational methods for real-time inference and reinforcement learning have driven applications in planetary rovers, underwater vehicles, autonomous cars and search-and-rescue platforms. The global significance of active SLAM lies in its capacity to enable resilient, efficient autonomy in GPS-denied or dynamically changing settings, supporting tasks as diverse as habitat monitoring, infrastructure inspection and exploration of extreme environments.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

A comprehensive decade-long review of active SLAM methodologies has synthesised developments in information-theoretic utility functions, optimal design criteria and collaborative multi-agent frameworks, highlighting trends in trajectory generation, uncertainty quantification and the integration of mapping with active perception. This survey identifies limitations in current evaluation metrics and proposes avenues for standardised benchmarks and real-world validation.

A model-free deep reinforcement learning approach has reframed active SLAM as a sequential decision problem in which reward functions embed classical information gains. Trained agents using a deep Q-network architecture learn to navigate and explore unknown maps with range sensors, demonstrating transferability to unseen environments and reduced computational overhead compared with traditional planning methods.

An active SLAM framework tailored to autonomous underwater vehicles combines an iterative closest point algorithm for pose-graph construction with a view-planner and entropy-based action selection. Trials in both simulation and field deployments show that bounding pose and map uncertainty through entropy minimisation yields more consistent reconstructions of complex subsea terrains, illustrating the robustness of active strategies in challenging, feature-sparse environments.

Active Simultaneous Localization and Mapping in Uncertain Environments publication trend

The graph below shows the total number of articles in active simultaneous localization and mapping in uncertain environments across all publications each year (not limited to Nature Index journals).

Technical terms

Belief state: The probabilistic distribution representing the robot’s estimate of its pose and the map of its environment.

Entropy: A measure of uncertainty or disorder within a probability distribution, used to quantify information content.

Frontier: The boundary between known and unknown regions in a map, commonly used to guide exploration targets.

Pose graph: A network structure in SLAM where nodes represent robot poses and edges encode spatial constraints derived from sensor measurements.

Information gain: The expected reduction in uncertainty achieved by executing a particular action or observation.

POMDP: A Partially Observable Markov Decision Process, a framework for decision-making under uncertainty in which states are not directly observable.

References

1. Active SLAM for Autonomous Underwater Exploration. Remote Sensing (2019).
2. Active SLAM: A Review on Last Decade. Sensors (2023).
3. A Deep Reinforcement Learning Approach for Active SLAM. Applied Sciences (2020).