UGV-Conditioned Multi-UAV Informative Planning on a Shared Exposure Belief

arXiv:2606.12306 · cs.RO · Submitted 2026-06-10 · Read on arXiv

Listen

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: Today's paper: "UGV-Conditioned Multi-UAV Informative Planning on a Shared Exposure Belief".

Dev: The gist:

Rosa: First, who's behind it and why it matters.

Paper summary: Rosa: So we're looking at this paper called "UGV-Conditioned Multi-UAV Informative Planning on a Shared Exposure Belief." Basically, it’s about helping a ground vehicle navigate safely when there are unknown threats around it, and they use a team of drones to help.

Dev: Right. The core idea here is that existing aerial systems just map things or cover an area, but this approach directs the sensing toward the areas that matter most for safety along the ground vehicle's route.

Taro: It’s about coordinating a team of UAVs to actively reduce risks for a ground vehicle moving through unknown threat zones.

Rosa: They introduce this key concept called a shared exposure belief, which gets updated online from aerial observations and is used jointly by both the UAV team and the ground vehicle.

Dev: That shared belief models something called the latent exposure rate lambda of x, which they model as a log-normal random variable. That gives them closed-form expressions for planning use.

Taro: So, this means both planners query this shared belief through a scalar risk-averse surrogate that penalizes cells with high uncertainty. Observing a cell actually reduces the effective uncertainty and thus lowers the risk cost for the ground vehicle route.

Rosa: The UGV then replans its route to minimize cumulative exposure under this current belief, which involves a per-cell dwell time defined as delta t j equals L j over v of x sub j.

Dev: The resulting per-cell cost increases both in slow terrain and in regions with high or uncertain exposure, and the UGV replans using an equation to minimize pi star g t, which is the path that minimizes h transpose pi plus beta E of pi t.

Taro: And they trigger this replanning whenever a UAV detects a new threat zone, meaning the whole system keeps updating in real time.

Rosa: Then there’s the part about UGV-Conditioned Informative Planning, where the planner is designed to collect aerial observations that most improve the UGV's exposure-aware routing.

Dev: They maximize an accumulated mission-conditioned reward under a motion budget bi, searching for tau star i which maximizes tau i x x in V of tau i R of x t s subject to C of tau i less than or equal to bi.

Paper summary: Taro: The reward function r of x t is constructed to reflect the operational objective, combining a base importance term with a priority gate that suppresses distant regions.

Rosa: Specifically, they have this final reward r of x t which includes gamma zero plus sigma g exp(x t) times the bar i of x t plus w lambda lambda bar(x t).

Dev: And to handle the multi-UAV setting without redundant sensing, they use region-conditioned spatial assignment based on a hierarchical region graph.

Taro: They assign Voronoi ownership to each UAV planner, and then they adjust the effective reward for UAV i by multiplying it by mi of x t which is based on whether x is in or outside its assigned region.

Rosa: This means setting vf equals zero gives hard coordination where each UAV perceives no reward outside its region, while setting vf greater than zero lets them respond to high-value areas near their boundaries.

Dev: The simulation experiments show that this approach reduces cumulative UGV exposure by thirty-eight percent compared to a system that doesn't account for hazard levels <ref:2606.12306#pg1,approach reduces cumulative UGV exposure by 38% compared to a system that>.

Taro: And they also reduce redundant aerial coverage from thirty-eight point eight percent down to three point seven percent under this multi-UAV coordination scheme.

Rosa: Exposure-aware planning using their shared exposure belief closes substantially more of the safety gap, which corresponds to a thirty-one percent reduction in mission objective and a thirty-eight percent reduction in cumulative exposure relative to terrain-semantic planning.

Dev: The gated reward achieves the lowest mission objective while preserving corridor coverage and priority threat detection.

Taro: Region-conditioned masking reduces redundant coverage from thirty-eight point eight percent down to five point zero percent, and then further down to three point seven percent under those specific conditions.

Rosa: But they also increased useful information efficiency from sixty-six point two percent up to more than ninety-two percent in those experiments.

Dev: The paper mentions that the gain is mainly caused by threat-related exposure, as ethreat reduces by about fifty-five percent compared to a thirty-three percent reduction in base exposure.

Taro: The authors also found that hard-regional coordination achieves the lowest mean objective, but it’s worth noting that priority-threat detection time worsens beyond four UAVs under hard-regional coordination, suggesting that increased Voronoi fragmentation at larger team sizes creates ownership boundaries that isolate individual UAVs from threats in neighboring regions.

Paper summary: Rosa: And they pointed out a trade-off there: hard masking improves spatial efficiency, but soft visibility helps preserve responsiveness near region boundaries.

Dev: So, to wrap up this paper on "UGV-Conditioned Multi-UAV Informative Planning on a Shared Exposure Belief," it shows that coupling aerial observations to ground-level route planning using a shared exposure belief improves UGV safety compared to terrain-semantic replanning alone.

Taro: It demonstrates that the UGV uses this belief to replan safer routes online, while the UAVs prioritize observations relevant to the current ground route, and they did this by using region-conditioned coordination.

Rosa: This approach shows that coupling aerial reconnaissance directly to exposure awareness improves UGV safety compared to terrain-semantic replanning alone.

Dev: The UGV-conditioned reward directs sensing effort toward mission-relevant regions, and the region-conditioned coordination substantially reduces redundant aerial coverage while maintaining responsiveness to relevant threat regions.

Rosa: So, looking at "UGV-Conditioned Multi-UAV Informative Planning on a Shared Exposure Belief" and its authors Oerlemans, Shi, and Popovic´. The main point is maintaining a shared exposure belief that couples aerial observations to ground-level route planning.

Dev: The UGV uses this belief to replan safer routes online, while the UAVs prioritize observations that are relevant to the current ground route.

Taro: This is interesting because they show that coupling aerial reconnaissance to a shared exposure belief improves UGV safety compared to terrain-semantic replanning alone.

Rosa: Their proposed UGV-conditioned reward directs sensing effort toward mission-relevant regions, and the region-conditioned coordination substantially reduces redundant aerial coverage while maintaining responsiveness to relevant threat regions.

Dev: The gain is mainly caused by threat-related exposure, as ethreat reduces by about fifty-five percent compared to a thirty-three percent reduction in base exposure.

Rosa: It’s a solid framework because it links aerial observations to route-relevant risk reduction rather than just semantic map updates.

Dev: The gain is mainly caused by threat-related exposure, as ethreat reduces by about fifty-five percent compared to a thirty-three percent reduction in base exposure.

Taro: The results confirm that coupling aerial reconnaissance to a shared exposure belief improves UGV safety compared to terrain-semantic replanning alone.

Conclusion: Rosa: So we've been looking at how this paper, "UGV-Conditioned Multi-UAV Informative Planning on a Shared Exposure Belief," works to make ground vehicles safer in dangerous areas, and now we're wrapping up with what it actually means.

Dev: Yeah, the main thing is that they couple aerial sensing directly with ground routing by keeping this shared exposure belief running between the UAV team and the UGV.

Taro: It means instead of just mapping things, the drones are looking exactly where they need to be looking based on how risky those areas are for the vehicle right now.

Rosa: They use this shared belief to let both systems replan online as threats pop up, which is different from just having a static map.

Dev: The results showed that this approach cuts cumulative exposure by thirty-eight percent compared to just using terrain-semantic replanning alone.

Taro: And they also managed to cut down on redundant aerial coverage significantly, from nearly forty percent down to about three point seven percent with the multi-UAV coordination setup.

Rosa: It points toward a way for aerial support not just to be about covering ground, but actively helping manage risk in real time.

Dev: The implication is that we can direct limited drone resources much more efficiently when we're trying to keep a vehicle out of harm's way.

Taro: But you have to watch how the team coordinates; they found that hard coordination works well for efficiency, but the team size really matters for threat detection time.

Rosa: Exactly. So this isn't just about having more drones; it’s about making those drones smart enough to prioritize what matters most for the vehicle’s immediate safety and path planning.

MAVLab, Faculty of Aerospace Engineering, TU Delft

cs.RO

Submitted: 2026-06-10

Updated: 2026-10-14

Comments: 8 pages, 6 figures

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 82/100

The gist: The gist: A UGV-conditioned multi-UAV informative planning framework couples aerial reconnaissance directly to exposure-aware ground navigation by maintaining a shared exposure belief that drives

Key concepts

Shared Exposure Belief: A common probabilistic map used by both the UGV and UAV team that tracks the risk of hazards along a route. It is updated continuously with new aerial observations, providing a consistent, shared understanding of where dangers are most likely to occur.
Exposure-Aware UGV Replanning: The ground vehicle finds safer routes by minimizing cumulative exposure. The cost function for routing increases based on terrain difficulty and the uncertainty or risk level of specific cells in the map, forcing the UGV to avoid high-risk areas.
UGV-Conditioned Informative Planning: The UAV team's planning is guided by a mission-conditioned reward. This reward directs sensing efforts toward regions that offer the greatest potential improvement for the UGV's safety, ensuring aerial observations are relevant to current ground navigation needs.
Region-Conditioned Coordination: To prevent multiple UAVs from sensing the same area repeatedly, spatial assignment is used based on Voronoi regions. This allows UAVs to divide the airspace efficiently, reducing redundant sensing while still allowing some overlap near boundaries for responsiveness.

Summary

The gist: A UGV-conditioned multi-UAV informative planning framework couples aerial reconnaissance directly to exposure-aware ground navigation by maintaining a shared exposure belief that drives both UAV sensing priorities and UGV route replanning.

Shared Exposure Belief

A shared probabilistic map is central to the framework, providing a consistent planning interface for both the UGV and the UAV team A shared probabilistic map is central to our framework, providing a consistent planning interface for both the UGV and the UAV team. This belief is updated online from aerial observations and used jointly by the UAV team and the ground vehicle. The latent exposure rate λ(x) is modeled as a log-normal random variable, which yields closed-form expressions for planning use. Both planners query the shared belief through a scalar risk-averse surrogate that penalises high-uncertainty cells. Observing a cell reduces σ eff ln and thus λrisk, lowering the UGV routing cost.

Exposure-Aware UGV Replanning

The goal is for the UGV to find a route that minimizes cumulative exposure under the current belief The per-step dwell time is defined as ∆tj = Lj/v(xj), where Lj ∈ δ, √2δ is the step length on the eight-connected grid and v(xj) is a terrain-dependent speed fraction. The resulting per-cell cost increases both in slow terrain and in regions with high or uncertain exposure. The UGV replans using Eq. 4 to minimize π⋆ g(t) = argmin π∈Π(xs,xg) h T(π) + β E(π, t), where E(π, t) = K X−1 j=0 λrisk(xj, t) ∆tj (5). Replanning is triggered whenever a UAV detects a new threat zone.

UGV-Conditioned Informative Planning

The planner is designed to collect aerial observations that most improve the UGV’s exposure-aware routing The planner maximizes the accumulated mission-conditioned reward under a motion budget Bi, searching for τ⋆ i = argmax τi X x∈V(τi) R(x, t) s.t. C(τi) ≤ Bi (6). The reward R(x, t) is constructed to reflect the operational objective It combines a base importance term I base(x, t) = (wpP(x, t) + wh) H(x, t), which is then passed through a priority gate I(x, t) = δf + (1 − δf) P(x, t) ηp · I base(x, t), where ηp > 0 controls how sharply the gate suppresses distant regions. The final reward is R(x, t) = γ0 + σg g exp(x, t) · ¯I(x, t) + wλ λ¯(x, t), where g exp(x, t) is a sigmoid gate that activates in high-exposure regions.

Multi-UAV Region-Conditioned Coordination

To prevent redundant sensing when multiple UAVs independently maximize the shared reward R(x, t), spatial ownership is assigned using Voronoi assignment on a hierarchical region graph The effective reward for UAV i is then Ri(x, t) = mi(x, t) R(x, t), where mi(x, t) = (1, x ∈ omegai(t), vf, x /∈ omegai(t)) (11). Setting vf = 0 yields hard coordination in which each UAV perceives no reward outside its region. Setting vf > 0 retains a fraction of foreign region reward, allowing UAVs to respond to high-value areas near their boundaries.

Experimental Results

Simulation experiments show that the approach reduces cumulative UGV exposure by 38% compared to a system that does not account for hazard levels It also reduces redundant aerial coverage from 38.8% to 3.7% under the multi-UAV coordination scheme. Exposure-aware planning using our shared exposure belief closes substantially more of the safety gap, corresponding to a 31% reduction in mission objective and a 38% reduction in cumulative exposure relative to terrain-semantic planning. The gated reward achieves the lowest mission objective while preserving corridor coverage and priority-threat detection. Region-conditioned masking reduces redundant coverage from 38.8% to 5.0% and 3.7%, respectively, while increasing useful information efficiency from 66.2% to more than 92%.

Conclusion

Our approach maintains a shared exposure belief that couples aerial observations to ground-level route planning The UGV uses this belief to replan safer routes online, while the UAVs prioritise observations that are relevant to the current ground route. We demonstrate that coupling aerial reconnaissance to a shared exposure belief improves UGV safety compared to terrain-semantic replanning alone. Our proposed UGV-conditioned reward directs sensing effort toward mission-relevant regions, and the region-conditioned coordination substantially reduces redundant aerial coverage while maintaining responsiveness to relevant threat regions. Future work will focus on extending the shared exposure belief with more realistic sensing and communication models. The results confirm that our proposed shared exposure belief links aerial observations to route-relevant risk reduction rather than only semantic map updates. The gain is mainly caused by threat-related exposure, as Ethreat reduces by about 55% compared to a 33% reduction in base exposure. Our gated reward achieves the lowest overall mission objective while maintaining nearly the same corridor coverage and priority-threat detection rate as weighted entropy. The bottom row of Fig. 5 shows representative trajectories for the evaluated coordination regimes, demonstrating that hard-regional coordination achieves the lowest mean objective. Hard-regional masking gives the best spatial efficiency and mission objective, but its strict ownership lowers the priority-threat detection rate to 45%. Soft-regional masking retains limited cross-region visibility between UAVs, which yields the highest TDR, while still keeping redundancy low. Hard-regional coordination achieves the lowest mean objective. The timing metrics show that more UAVs do not automatically improve all mission-relevant outcomes. Notably, priority-threat detection time worsens beyond four UAVs under hard-regional coordination, suggesting that increased Voronoi fragmentation at larger team sizes creates ownership boundaries that isolate individual UAVs from threats in neighbouring regions. This reinforces the trade-off that hard masking improves spatial efficiency, whereas soft visibility helps preserve responsiveness near region boundaries.

--- Page 1 ---

UGV-Conditioned Multi-UAV Informative Planning on a Shared Exposure Belief Lars Oerlemans, Moji Shi, and Marija Popovic´ Abstract— Safe ground navigation in large, threat-augmented environments requires aerial support that actively reduces the risks that a ground vehicle faces along its route Existing aerial reconnaissance systems focus on mapping or covering the environment, but do not direct sensing toward regions that are most relevant for ground vehicle safety In this paper, we address the problem of coordinating a team of unmanned aerial vehicles (UAVs) to improve the safety of an unmanned ground vehicle (UGV) navigating through unknown threat zones A key aspect of our approach is a shared exposure belief that is updated online from aerial observations and used jointly by the UAV team and the ground vehicle This enables us to direct aerial sensing towards route-relevant regions while allowing the UGV to replan around newly revealed threats We coordinate the UAV team through spatial region assignment to avoid redundant sensing Simulation experiments show that our approach reduces cumulative UGV exposure by 38% compared to a system that does not account for hazard levels, and reduces redundant aerial coverage from 38.8% to 3.7% under our multi-UAV coordination scheme I.

Improvements for AI systems

A shared exposure belief that is updated online from aerial observations and used jointly by the UAV team and the ground vehicle enables a system to simultaneously direct aerial sensing towards route-relevant regions while allowing ground vehicle replanning around newly revealed threats.
The UGV planner can find a path minimizing h T(π) + β E(π, t) where E is the cumulative exposure under the risk-averse surrogate, ensuring routes are optimized for safety rather than just traversal time.
The UAV planning utility is defined as maximizing a mission-conditioned reward R(x, t), which combines corridor relevance with exposure and entropy to direct sensing effort toward regions that most improve ground-level safety.
Region-conditioned coordination assigns spatial ownership to each agent to reduce redundant sensing while matching responsiveness to newly relevant regions, preventing the convergence of multiple UAVs onto the same high-reward areas.

Abstract

Safe ground navigation in large, threat-augmented environments requires aerial support that actively reduces the risks that a ground vehicle faces along its route. Existing aerial reconnaissance systems focus on mapping or covering the environment, but do not direct sensing toward regions that are most relevant for ground vehicle safety. In this paper, we address the problem of coordinating a team of unmanned aerial vehicles (UAVs) to improve the safety of an unmanned ground vehicle (UGV) navigating through unknown threat zones. A key aspect of our approach is a risk-averse shared exposure belief for UGV routing and a mission-conditioned UAV reward that directs sensing toward uncertain, exposure-relevant regions of the current UGV corridor. UAV observations update the shared belief online, allowing the UGV to replan around newly revealed threats while the evolving route in turn reshapes UAV sensing priorities. For multiple UAVs, we use region-conditioned spatial assignment to reduce redundant sensing. Simulation experiments show that our approach reduces cumulative UGV exposure by 38% compared with terrain-semantic replanning, and reduces redundant aerial coverage from 38.8% to 3.7% under our multi-UAV coordination scheme.

Sources

Dynamic UGV-UAV Cooperative Path Planning in Uncertain Environments