Odor Source Localization in Robotic Systems

Odor source localization is the process by which autonomous robots detect, track and ultimately pinpoint the origin of chemical emissions in complex environments. Ground and aerial platforms equipped with chemical sensors navigate turbulent airflows, sparse cue distributions and variable wind conditions to assemble spatial and temporal data on airborne chemicals. Key challenges include the intermittent nature of odour plumes, slow sensor dynamics, environmental obstacles and real-time decision making under uncertainty. To overcome these hurdles, researchers have devised reactive strategies inspired by insect behaviour, probabilistic search methods that balance exploration and exploitation, and machine-learning approaches that infer plume structure or predict source proximity. Applications span environmental monitoring, detection of hazardous gas leaks in industrial plants or disaster sites, search-and-rescue operations in collapsed structures and precision agriculture. Progress has been driven by advances in compact gas sensors, real-world and simulated plume modelling, adaptive control algorithms and integration of multimodal data streams such as wind velocity and vision. Contemporary systems demonstrate increasing autonomy, robustness to wind variability and the capacity to build three-dimensional gas concentration maps in real time.

Research from Nature Portfolio

Recent studies have taken inspiration from biological olfaction to develop artificial agents that track odour plumes under changing wind conditions. By training recurrent neural networks with deep reinforcement learning in simulated turbulent flow fields, agents have emerged that mimic insect surge-and-cast behaviours, dynamically compute task-relevant variables and maintain memory of plume encounters. Analyses of network activity reveal distinct neural dynamics underpinning successful tracking and generate testable hypotheses on the memory demands of plume following. These insights bridge computational neuroscience and autonomous control, offering guidelines for the design of compact algorithms capable of adapting to wind direction shifts and intermittent odour cues.

Research from all publishers

Engineers have miniaturised gas-sensing drones to create nano-aerial vehicles that autonomously map and localise gas sources in confined or hazardous settings. Lightweight quadcopters bearing metal oxide semiconductor sensors perform sweeping flight patterns, extract transient plume features in real time and reconstruct three-dimensional concentration fields to identify leak points with metre-level accuracy. This approach demonstrates practical utility for search-and-rescue missions in collapsed buildings and industrial inspections.

Alternative strategies employ joint Bayesian estimation and path-planning on ground robots equipped with low-cost gas sensors. By recursively updating posterior distributions of source location and strength and selecting waypoints that maximise expected information gain, these systems locate simulated hazardous releases in turbulent indoor conditions. Experimental trials confirm that such information-driven navigation accelerates source term estimation compared with naive gradient following.

Entropy-based search frameworks, known as entrotaxis, represent another line of development. These methods apply sequential Monte Carlo to maintain probability maps of source location, then steer robots to regions that reduce overall uncertainty most efficiently. Comparative studies show entrotaxis can outperform classic infotaxis in sparse-cue or highly intermittent environments, achieving faster convergence and reduced computational overhead during decision making.

Odor Source Localization in Robotic Systems publication trend

The graph below shows the total number of articles in odor source localization in robotic systems across all publications each year (not limited to Nature Index journals).

Technical terms

Odour plume: A spatially and temporally discontinuous distribution of airborne chemical cues shaped by turbulent flow, wind direction and environmental obstacles.

Deep reinforcement learning: A machine-learning paradigm in which artificial neural networks learn to make sequential decisions by maximising cumulative rewards through interaction with a simulated or real environment.

Metal oxide semiconductor (MOX) sensor: A chemical transducer that detects volatile compounds via changes in electrical resistance of a metal oxide surface when exposed to gas, valued for sensitivity and low cost but limited by slow recovery times.

Bayesian inference: A statistical framework for updating the probability distribution of unknown parameters (such as source location) based on new sensor measurements and prior beliefs.

Infotaxis: A search algorithm that directs a robot to locations predicted to yield the greatest information gain about the source position, balancing exploration of new areas with exploitation of known cues.

References

1. Emergent behaviour and neural dynamics in artificial agents tracking odour plumes. Nature Machine Intelligence (2023).
2. Smelling Nano Aerial Vehicle for Gas Source Localization and Mapping. Sensors (2019).
3. Entrotaxis as a strategy for autonomous search and source reconstruction in turbulent conditions. Information Fusion (2018).
4. Information-Based Search for an Atmospheric Release Using a Mobile Robot: Algorithm and Experiments. IEEE Transactions on Control Systems Technology (2018).
5. Reactive Searching and Infotaxis in Odor Source Localization. PLOS Computational Biology (2014).
6. Exploiting plume structure to decode gas source distance using metal-oxide gas sensors. Sensors and Actuators B Chemical (2016).