← Paper Digest

Disturbance-Aware Informative Planning in Spatiotemporal Environments

Kaleb Ben Naveed, Haejoon Lee, Anouck Girard, Dimitra Panagou

November 2026

TL;DR A two-stage planner for persistent monitoring first measures unknown wind only along the routes that matter for monitoring, then optimizes trajectories under the learned wind. In four simulated scenarios it spends 25.0-52.3% less time learning the wind than mapping it across the whole workspace.

What they do

A robot that persistently monitors a changing quantity, such as temperature or water quality, loses information over time and has to keep revisiting regions. Unknown, spatially varying wind pushes it off its planned path, so trajectories optimized with nominal dynamics collect less information than predicted. This work learns the wind only where it affects monitoring, then plans with the learned wind.

How it works

  • Monitoring-relevant paths. Predict where clarity (a normalized entropy measure) will fall below target over the 15 s horizon, and generate several geometrically distinct collision-free paths toward those regions, scored by the deficits observable along them.
  • Targeted wind sampling. Wind is modeled with POAM, an online sparse Gaussian process. The robot measures where the expected reduction of wind uncertainty near high-scoring paths is largest, with a travel-distance penalty, until the weighted uncertainty drops below a threshold.
  • Disturbance-aware planning. Model-Based Diffusion, a sampling-based optimizer, refines the geometric paths. Each candidate is scored by rolling it out under wind fields sampled from the GP, with a robust control-barrier-function safety filter applied inside the rollouts.

Results

All in simulation: a 2D double integrator with a wind force model, four scenarios, one run each.

  • Dedicated wind learning took 180, 200, 124 and 102 s, against 240, 370, 244 and 214 s for learning wind over the whole workspace.
  • Scenario 1: time-averaged clarity deficit of 0.166 against 0.224 for disturbance-aware planning without a learning stage, which never reached the far monitoring region; the global-learning baseline hit an obstacle at 399 s.
  • Scenario 3: 0.0418 against 0.0540 for global learning (22.6% lower). Median replanning time was 1.64 s.

Limitations

One run per scenario, a point-mass robot, time-invariant wind and no hardware. In the largely calm Scenario 4 a zero-wind planner did better (0.180 against 0.189), so dedicated learning can be wasted. Infeasible safety-filter steps occurred, and the authors state the results do not establish guaranteed collision avoidance.

Figures

From Figure 1: Overview of the two stages, monitoring-informed disturbance estimation followed by disturbance-aware i...
From Figure 1: Overview of the two stages, monitoring-informed disturbance estimation followed by disturbance-aware informative planning.
From Figure 2: Monitoring across the four scenarios. Top: environments and trajectories of the proposed method; botto...
From Figure 2: Monitoring across the four scenarios. Top: environments and trajectories of the proposed method; bottom: mean clarity-deficit histories for all methods.
From Figure 4: Trajectories of the proposed method and the baselines, one row per scenario (Scenarios 1-4; the embedd...
From Figure 4: Trajectories of the proposed method and the baselines, one row per scenario (Scenarios 1-4; the embedded labels of the last two rows are swapped). Runs that ended in contact were terminated.
Abstract

Autonomous robots are widely used for persistent monitoring of spatiotemporal environments, where quantities of interest (QoI) change across space and time and require repeated measurements. In practice, robot motion can be affected by external disturbances that vary across the environment. When such external disturbances are not considered in trajectory optimization, it can lead to inaccurate predictions of robot motion and, consequently, the information acquired about the QoI during execution. However, estimating disturbances throughout the environment can be inefficient, as disturbances in regions with little overlap with monitoring trajectories have limited impact on monitoring performance. We therefore propose a two-stage informative planning framework for unknown, spatially varying but time-invariant disturbances, combining informed disturbance sampling with disturbance-aware trajectory optimization. First, we identify where QoI measurements are needed and generate alternative paths to these locations using the environment geometry. We then prioritize disturbance measurements based on their expected reduction of disturbance uncertainty along these paths. Second, rather than optimizing information acquisition based on nominal robot motion, we incorporate the learned disturbance belief into trajectory optimization to account for its effects on robot motion and information acquisition. Across four simulation scenarios, the proposed method reduces disturbance-learning time by 25.0--52.3 percent compared with environment-wide estimation and identifies paths that appear costly under nominal dynamics but become preferable when disturbances are considered.

Summaries are written automatically from each paper's text. Figures and videos come from the paper and its project page; all credit goes to the authors.