XIT: Exploration and Exploitation Informed Trees for Active Gas Distribution Mapping in Unknown Environments

Mal Fazliu; Matthew Coombes; Sen Wang; Cunjia Liu

cs.ROFeb 14, 2026

XIT: Exploration and Exploitation Informed Trees for Active Gas Distribution Mapping in Unknown Environments

Authors: Mal Fazliu, Matthew Coombes, Sen Wang, Cunjia Liu

Organizations: Department of Aeronautical and Automotive Engineering, Loughborough University, Leicestershire, UK, LE11 3TU · Department of Electrical and Electronic Engineering, Faculty of Engineering, Imperial College London, South Kensington Campus, London, UK, SW7 2AZ

Abstract

Mobile robotic gas distribution mapping (GDM) provides critical situational awareness during emergency responses to hazardous gas releases. However, most systems still rely on teleoperation, limiting scalability and response speed. Autonomous active GDM is challenging in unknown and cluttered environments, because the robot must simultaneously explore traversable space, map the environment, and infer the gas distribution belief from sparse chemical measurements. We address this by formulating active GDM as a next-best-trajectory informative path planning (IPP) problem and develop XIT (Exploration and Exploitation Informed Trees), a sampling-based planner that grows a single shared tree toward multiple frontier goals, coupling exploration of unresolved regions with informative gas sampling en route. XIT draws a batch of samples from an Upper Confidence Bound (UCB) information field derived from the current gas posterior and expands the tree using a cost that trades off travel effort against information acquisition. To enable plume-aware exploration, we introduce the gas frontier concept, defined as unobserved regions adjacent to high gas concentrations, and propose the Wavefront Gas Frontier Detection (WGFD) algorithm for their identification. High-fidelity simulations demonstrate the benefits of XIT in terms of GDM quality and efficiency, and a scaled real-world experiment demonstrates its practicality. Although developed for active GDM, XIT is applicable to other robotic information-gathering tasks in unknown environments that face the exploration and exploitation trade-off.

Figures & tables

Fig. 1: Overview of the XIT planner for active GDM. The figure illustrates a robot constructing a gas distribution map from concentration measurements in real time and planning gas-exploitative, distance-aware trajectories toward both gas frontiers and environmental frontiers. The three candidate paths generated by XIT are shown in grey, with the next-best trajectory (NBT) highlighted in green, the path with the highest expected information value, balancing gas dispersion inference and spatial exploration where possible.

Panel of Figure 2 — Fig. 2: System overview of the proposed active GDM framework. On the left (a) is a general system architecture for the mobile robotic application. Key functional components in the compute unit include localisation and mapping for navigation and occupancy information (summarised as the SLAM module), the GDM function, and the IPP module, which selects the NBT . The IPP process is expanded in (b), where we also illustrate a gas-polluted environment and the corresponding mean and uncertainty maps produced during GDM . The concept of gas frontiers is visualised using yellow markers along the boundary of the estimated gas plume, while conventional frontiers outlining the unknown occupancy map are in red. Both types are used by XIT to steer and terminate tree expansion in exploration-rich regions, while a UCB -based cost function, derived from the current gas model, guides paths through inferred high-value gas areas. A separate example is included to illustrate the characteristic trajectories and tree structure generated by XIT .

Fig. 3: Effect of the distance weight $\alpha$ on an isolated XIT trajectory to a single goal, for $\alpha=0$ (a), $0.5$ (b), and $5$ (c). All panels use the same scenario, start, goal, and sample set, so the differing routes are due to $\alpha$ alone. A large sample budget with full graph-optimal expansion is used to show the converged routes: at $\alpha=0$ the cost is purely the information penalty and the path detours through high- UCB regions, whereas larger $\alpha$ favours the shorter geometric route. Colour shows the UCB information field; grey lines are tree edges and orange the selected trajectory.

Fig. 4: Example trajectories produced by XIT in three synthetic gas-polluted scenarios (a)–(c). The left column shows each scenario, including the UCB information field, obstacles, start state, and frontier-derived goals. The right column shows the single XIT tree (grey), grown concurrently toward all goals, together with the trajectory recovered to each goal. The trees use $N=150$ informed samples and $\alpha=0$ , so the explicit Euclidean-distance term is absent and the selected trajectories preferentially pass through high- UCB regions while progressing toward their respective goals.

Fig. 5: Cell-level depiction of gas frontiers and their role in path planning within an active GDM task over three sequential time steps. (a) The robot enters from the bottom left and finds itself within a critical gas region above $\tau_{\text{gas}}$ based on its updated belief. Two gas frontiers are identified, and one is selected as the NBT . (b) Upon reaching the gas frontier, the updated belief reveals that the area is no longer a priority as no new gas frontiers emerged based on the estimate, suggesting a less concentrated region. It now targets the other previously identified gas frontier from (a), which has remained unresolved. (c) Resolving the second gas frontier reveals a high-concentration zone that generates a new gas frontier, which can be explored next to continue the adaptive plume probing process.

Fig. 6: Goal-averaged graph-optimal trajectory cost for the three scenarios in Fig. 4 , against sample budget

N

. UCB -guided mixture sampling is compared with uniform sampling using

\alpha=0

. Curves show the mean over the runs (of 50 random seeds) in which all three goal regions were reached; shaded bands indicate 95% confidence intervals, and labels give the success rate where below 100%. Both strategies approach the same low-cost regime as

N

increases, whereas UCB -guided sampling provides an advantage when the sample size is limited, most visibly in (a) and (b).

Fig. 7: Mean gas-map RMSE (solid, left axis) and mean travel effort (dashed, right axis) versus the number of goals visited (one goal per planning cycle), over 50 seeded missions in the three scenarios of Fig. 4 ((a)–(c)). Smaller

\alpha

invests greater travel effort per goal and yields consistently lower mapping error.

Fig. 8: Goal-averaged trajectory cost over runs reaching all

M

goals (50 seeds; band: 95% confidence interval) versus wall-clock planning time, for XIT (graph-optimal and feasibility-first) and a sequential BIT* baseline (

\alpha=0

). Rows (a)–(c) are the three scenarios of Fig. 4 ; columns are

M=1,3,6,9

goals. Each marker is a different sample budget

N

, not the progress of a single run. Labels give the fraction of runs reaching all

M

goals where below 100%.

Fig. 9: High-fidelity simulation scenarios: (a) the cluttered industrial indoor layout and (b) the warehouse, both with multi-source acetone dispersion. White lines denote walls and magenta blocks storage obstacles; green particles visualise the dispersion field; crosses mark the labelled acetone sources and robot start.

Fig. 10: Ground-truth gas distributions of (a) the industrial scenario and (b) the warehouse scenario. Crosses mark the acetone sources (S1–S3, in the order of Table I ), white regions are obstacles, and the critical gas regions used in the evaluation are outlined in red.

Environment & Gas Sources
Parameter	Value (industrial / warehouse)
Environment size [m]	$34\times 43$ / $80\times 40$
Release rates S1, S2, S3 [ppm/s]	1750, 1250, 1000
Filament growth rate [cm 2 /s]	15
Wind field [m/s]	0.3 / 0.1 (steady)
Robot Configuration & Sensing

TABLE I: Key Simulation Parameters.

Parameter	Value (industrial / warehouse)
Batch size ( $N$ ) [samples]	300 / 400
Maximum goal regions	5
Nearest samples per goal ( $k_{n}$ )	3
Uniform sampling component	20%

TABLE II: XIT Planner Configuration Parameters.

Fig. 11: Illustrative run of the active GDM process in simulation (industrial scenario) using XIT -GFF with UCB cost. Only the selected NBT is shown in each phase for clarity. The normalised posterior mean gas estimate is overlaid using the viridis colormap (yellow indicates higher concentration). (a) Initial structural exploration with occupancy frontiers. (b) Gas frontier detection triggers targeted probing of critical gas boundaries. (c–e) Alternation between occupancy exploration and gas-structure exploration as new critical regions are encountered. (f) Example of a gas-exploitative path that traverses high-concentration regions rather than taking the shortest route.

Fig. 12: Performance comparison over time in the industrial (left, 300 s missions) and warehouse (right, 600 s missions) scenarios, in terms of RMSE (top), total differential entropy (middle), and map completeness (bottom), for different goal-selection policies and cost formulations. Shaded regions indicate 95% confidence intervals for XIT -GFF with UCB (best performing) and RRT* Frontier Exploration (baseline).

Industrial scenario
Method	Cost	RMSE T [ppm]	RMSE Reduction [%]	$H_{T}$	Entropy Reduction [%]
RRT* Frontier Exploration	Euclidean	$16.7\pm 2.6$	$0.0$	$6652\pm 524$	$0.0$
XIT-F	UCB	$16.1\pm 2.8$	$3.3$	$6395\pm 862$	$3.9$
XIT-FGF	Euclidean	$16.4\pm 2.7$	$1.5$	$6488\pm 728$	$2.5$
XIT-FGF	UCB	$15.8\pm 3.1$	$5.4$	$6032\pm 987$	$9.3$
XIT-GFF	Euclidean	$15.2\pm 2.2$	$8.8$	$5838\pm 578$	$12.2$

TABLE III: Final GDM Performance at Mission End (

T{=}300

s Industrial,

T{=}600

s Warehouse; Mean

\pm

Std Over 20 Trials).

Panel of Figure 16 — Fig. 13: Experimental setup. (a) Top-down view of the 5.0 m × 3.7 m arena showing airflow directions from fans (dashed arrows) and CO 2 release nozzles (Source 1 and Source 2). Note that the two fans on the left are more powerful than the single and smaller fan on the right, creating asymmetric dispersion patterns. (b) BOC 8500 regulator attached to the pressurised CO 2 cylinder used for controlled gas release. (c) TurtleBot3 platform equipped with a COZIR-A CO 2 sensor and RP-LiDAR for gas mapping and navigation.

Fig. 14: Progression of the real-world experiment using the XIT planner in the scaled indoor gas-leak environment. Each row shows the top-down view of the arena (left) with the normalised gas mean map projected on the floor, where red, yellow, and purple lines indicate frontiers, gas frontiers, and the NBT , respectively. The corresponding XIT paths and UCB information field are shown on the right with obstacle inflation in grey. From top to bottom: XIT begins in a pure exploration phase, as the starting location lies outside the critical gas region, proceeds through gas-exploration phases around both sources (the second shown in the middle), and concludes with a gas-exploitative trajectory that passes downwind through both gas regions. A video of the experiment can be found in the supplementary material.

Fig. 15: Final mean gas concentration map obtained in the real-world experiment using the XIT planner within the active GDM framework. Source locations and airflow directions (dashed arrows) are indicated as in Fig. 13 a.

Informative Path Planning

Informative path planning chooses robot routes and sensing actions that gather useful information while respecting travel, energy, time, and safety constraints.

See all 37 papers

Sep 30, 2026

GPU-Accelerated Path-Dependent Marginal Information Gain for Autonomous Exploration

Autonomous exploration demands that robots continuously evaluate candidate viewpoints based on their expected information gain and execution cost. Sampling-based planners estimate this gain by volumetric raycasting and, due to its computational cost, evaluate candidates under an assumption of mutual independence, ignoring the overlap between viewpoints along the same path. This work presents a GPU-accelerated method for computing path-dependent marginal information gain, where instead of storing and merging the observed unknown voxels along each candidate path, previous observations are represented using depth buffers. Candidate rays are projected into the depth buffers of their ancestors to identify observation overlap and exclude regions expected to be observed. The planning tree is evaluated in depth order to maintain the dependency between viewpoints and their optimized yaws, while candidate nodes and rays at each level are processed in parallel on the GPU. The proposed method stays within 5-10% of the exact marginal gain computed using voxel hash maps, with speed-ups of up to 118x on a desktop GPU and 28x on an NVIDIA Jetson Orin NX. The method was integrated into two sampling-based exploration planners and evaluated in three simulation environments, where marginal gain reduced the time to 95% coverage in five of the six evaluated planner-environment combinations. Real-world experiments also showed a 30% reduction in the time to 95% coverage, as well as earlier exploration termination times.

João Félix Mendes, Rodrigo Ventura, Meysam Basiri

Institute for Systems and Robotics, Department of Electrical and Computer Engineering, Instituto Superior Técnico, Lisboa, Portugal

GPU Acceleration Robot Navigation Informative Path Planning Sampling-Based Motion Planning

Ask PDF

Sep 17, 2026

SmellDiffusion: Diffusion-Based Quadruped Navigation with Olfactory Scene Graphs

A robot sent to a named gas leak must preserve gas identity, estimate the source, and navigate to the resulting goal. We present SmellDiffusion, a simulation pipeline that represents species-specific gas zones in an open-vocabulary olfactory scene graph and shares the selected goal between classical and diffusion planners. Its key components are a peak-local geometric gate for selective source correction and diffusion-based, gas-guided trajectory generation. Among 424 unique source-wind configurations in solved flow, 28 have a concentration peak displaced more than 0.5m from the source. A source-independent geometric gate, calibrated only on the training split and evaluated at the observed peak, detects 9 of 10 held-out displacements at 0.64 precision. Gating a precomputed forward-matching correction reduces mean error on the displaced cases from 1.468m to 0.592m (60%), using matching for only 14/204 cases. All-case mean error falls from 0.205m to 0.180m. All planners receive the same scene-graph source estimate as their goal. In a controlled comparison, best-of-ten diffusion achieves mean gas exposure comparable to gas-guided A* (0.0476 versus 0.0455). A single diffusion proposal takes 41.7ms, compared with 72.3ms for gas-guided A*, although best-of-ten sequential sampling increases total runtime. Plain A* also reaches the same goal and remains the fastest and shortest-path method. Six matched Gazebo runs give mean robot-to-source errors of 0.39m for A* and 0.31m for diffusion.

Faith Ogunwoye, Iana Zhura, Hajira Amjad +5

Skolkovo Institute of Science and Technology, Moscow, Russia

Trajectory Generation Diffusion-Based Planning Robot Navigation

Ask PDF

Sep 19, 2026

MIRA-PRM: Mission-Informed Reusable Roadmap Planning for Mobile Gas-Sensing Inspection

Mobile gas-sensing inspection requires a mobile platform to reliably reach ordered sampling poses. We present MIRA-PRM, a mission-informed probabilistic roadmap that combines geometry-, mission-flow-, and inspection-conditioned sampling with locally adaptive connectivity and evidence-gated refinement. Eight development campaigns comprising 142,448 planner runs evaluated mission scaling, repeated-query adaptation, finite node budgets, target distributions, local map repair, ablation, cross-family comparison, and parameter sensitivity. MIRA-PRM maintained 100% seed-level success as ordered targets increased from two to eight, and achieved 97.67% seed-level success under a 40-node ceiling, compared with 67.17% for PRM and 86.50% for PRM*. Local repair reduced event time by 79.9-81.9% and node additions by 90.8-94.3% relative to rebuilding. On a separate frozen holdout of 12 maps, 48 ordered missions, and three seeds per map-mission unit, MIRA-PRM produced 96/144 validated successes, versus 82/144 for PRM, 89/144 for PRM*, and 70/144 for LE-HG-PRM. On paired common-success runs, MIRA-PRM reduced wall time by 16.89% relative to PRM* and 81.01% relative to LE-HG-PRM, but was 52.23% slower than PRM; paths were 0.54-1.86% longer. All validated paths passed independent full-segment checks and clearance non-inferiority. The results support a reliability-oriented trade-off in structured simulations and motivate external-map and physical mobile-sensing validation.

Gongsen Wang, Siyuan Wang, Xinyuan Wang +4

State Key Laboratory of Transducer Technology, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100190, China · School of Electronic, Electrical and Communication Engineering, University of Chinese Academy of Sciences, Beijing 100049, China · School of Electronic and Optical Engineering,Zijin College,Nanjing University of Science and Technology,Nanjing 210023, China +4

Roadmap-Based Motion Planning Sampling-Based Motion Planning Robotic Inspection

Ask PDF

Abstract

Informative Path Planning

Explore similar work

GPU-Accelerated Path-Dependent Marginal Information Gain for Autonomous Exploration

SmellDiffusion: Diffusion-Based Quadruped Navigation with Olfactory Scene Graphs

MIRA-PRM: Mission-Informed Reusable Roadmap Planning for Mobile Gas-Sensing Inspection