Pith. sign in

REVIEW 4 major objections 4 minor 36 references

Semantically-driven Deep Reinforcement Learning for Inspection Path Planning

T0 review · 4 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that a single end-to-end reinforcement-learned policy can simultaneously search for, fly to, and visually inspect objects of interest in unknown environments, using only local onboard observations and no prior map.

desk verdict Solid end-to-end RL inspection planner with real deployment; main flaw is overclaiming 'previously unseen semantics' when real masks are hand-tuned HSV thresholds. read the letter →

arxiv 2505.14443 v1 pith:RCGQUYJD submitted 2025-05-20 cs.RO

Grzegorz Malczyk , Mihir Kulkarni , Kostas Alexis This is my paper · ORCID
classification cs.RO
keywords inspectionpathplanningsemanticdeepreinforcementlearningaerialrobotsinformativesim2realtransfervisualcoveragePOMDP
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper aims to show that a single end-to-end reinforcement-learned policy can carry out a targeted inspection mission in an unknown environment: find an object of interest, approach it, and visually cover its surface, all while avoiding obstacles and using only local sensor data. This matters because inspection missions usually want a few sparse objects inspected, not whole-space coverage, and existing planners either cover everything or need environment-specific tuning or explicit semantic models. The proposed policy takes a masked depth image, a local occupancy grid around the robot, and a map of where the robot has already been, and outputs velocity and yaw-rate commands directly. The authors report that the same policy, trained only on primitive geometric shapes in simulation, inspects new concave structures and multiple never-seen objects in real deployments, including a mission with 96.2 percent average coverage of feasible surfaces.

What carries the argument

The load-bearing mechanism is a neural policy that fuses two local 3D maps with one masked depth image. A three-dimensional residual encoder processes the ego-centric occupancy grid, a small cube of occupied, free, and unknown cells centered on the robot; a second encoder processes the spatial visit score map, a grid in which each cell carries the Shannon entropy contribution $m^h_{i,j,k} = -p_{i,j,k}\log p_{i,j,k}$ of the fraction of visits the robot has made to that cell. A two-dimensional encoder processes the depth image multiplied element-wise by the semantic mask, $D_t \circ S_t$, so the policy sees range to the target but not the surrounding scene. The encoded features are concatenated with the robot state and passed through a multi-layer perceptron and a gated recurrent unit, which outputs a four-dimensional command of linear velocities and yaw rate. This design lets the policy memorize where it has looked, avoid already-visited viewpoints, and keep searching when the target is not yet in view.

What would settle it

Inspect the same three real objects under illumination that shifts their apparent color outside the fixed threshold ranges, or drive the policy with masks from a learned class segmenter instead of color thresholds; if coverage drops sharply or the search phase fails, the demonstrated capability is color-following rather than semantic inspection.

Watch

Extended reading notes

Core claim

The central claim is that semantics-aware inspection planning does not need a global map, a predefined semantic model, or a separate path planner. The authors state that they contribute the first end-to-end reinforcement-learning inspection planning framework that simultaneously performs semantic visual inspection and collision-free navigation from local observations. The policy is trained as a partially observable Markov decision process with a one-step reward that sums face-mesh coverage, a search bonus that decays with local re-visitation, and a collision penalty. At deployment it treats the masked depth image as the only semantic channel, so object shape, size, color, and class are not part of the state. On the strength of environment randomization during training and ablations showing that the visit-history map and recurrent memory add 22.3 and 14.8 percentage points of coverage respectively, the authors conclude the approach generalizes across obstacle layouts and crosses the simulation-to-real gap.

Load-bearing premise

The real-world claim of handling previously unseen semantics assumes that the color-threshold masks used to mark test objects keep working under new lighting and backgrounds; if the color segmentation fails, the policy receives no semantic signal at all.

Editorial extensions

If this is right

  • If the approach holds up, targeted inspection planning in unknown environments can skip global mapping and explicit semantic models.
  • A single policy can switch between multiple targets by changing the semantic mask, as demonstrated in the three-object laboratory mission.
  • Object shapes never seen in training, including concave bracket toes and T-shaped pipes, are inspectable at near-full feasible coverage.
  • The roughly 100 ms onboard inference time suggests the approach fits the compute budget of small aerial platforms even without optimization.
  • Continuous velocity and yaw-rate commands let the policy maneuver more tightly than waypoint-following baseline planners, yielding faster coverage.
  • Ablations show the visit-history map and the recurrent memory are both load-bearing: removing either one lowers coverage substantially.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's real-world semantics are fixed color-threshold masks on objects the experimenters selected and colored, so the demonstrated claim is generalization to new geometries, not yet to new semantic classes; a learned segmenter would be the natural next test.
  • Because the policy consumes only a masked depth image, the same network could be coupled to any upstream object detector or segmenter that outputs binary masks without retraining; that extension is untested.
  • The visit-entropy map should be stress-tested in environments where a target is visible only through a single narrow gap, since the entropy bonus may encourage the robot to leave a promising viewpoint before coverage is complete.
  • The face-reward uses a Gaussian in view distance, which predicts that coverage quality should degrade smoothly as the required inspection distance moves away from the range used in training; this relationship is testable.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 4 minor

Summary. The paper proposes a deep reinforcement learning policy for semantics-aware inspection path planning. The policy takes as input the current state, the previous action, a semantically masked depth image, an ego-centric local occupancy map, and a spatial visit score map, and outputs velocity and yaw-rate commands. The reward combines a face-mesh coverage term, a semantic-search exploration term, and a collision penalty. Training is performed in the Aerial Gym Simulator with APPO on primitive-shaped objects. The method is evaluated in simulation against FUEL, GVI, and SWAP in a water ballast tank and a chemical plant scenario, and is deployed on a real quadrotor inspecting a spherical light chamber and a multi-object industrial scene. The paper claims to be the first semantics-aware end-to-end RL inspection planning framework and to generalize to previously unseen semantics and unknown environments.

Significance. If the claims are supported, the paper would make a useful contribution: a compact learned policy that performs targeted object inspection and collision-free navigation using purely local observations, with demonstrated transfer from simulation to a real quadrotor. The paper has several concrete strengths: it is open-sourced, it compares against established baselines rather than only self-baselines, it evaluates on real-world mesh reconstruction, and the ablation studies isolate the role of the SVS map and the GRU memory. These features give the central planner idea independent grounding; the core contribution is not circular. However, the real-world evidence currently bears mainly on geometry and dynamics transfer with known-color objects, not on transfer of semantic perception to unseen semantics, and the headline real-world coverage number is affected by a post-hoc exclusion. These are load-bearing gaps between the claims and the presented validation.

major comments (4)
  1. [Section IV-C, Abstract, Section I-A] The real-world generalization claim is not supported by the evidence. The deployed segmentation is obtained by "creating a binary mask based on specific colour ranges in the HSV colour space" for a red T-shaped pipe, a blue barrel, and a yellow hydrant selected for the experiment, while training uses perfect simulator segmentation masks. This demonstrates inspection of known-color objects, not "previously unseen semantics" or a policy independent of "predefined sets of semantic classes" as stated in the abstract and Section I-A. Since the masked depth D_t(S_t) is the only object-specific observation, the policy's real-world behavior is directly conditioned on the reliability of these hand-tuned color thresholds, and no mask accuracy, lighting variation, or appearance variation is reported. The authors should either provide a semantic segmentation method that is evaluated on unseen classes in new environments, or revise the abstract and Section I-A to state that the real-world validation uses color-based masks for known target colors.
  2. [Section IV-C, multiple-object experiment] The headline average of 96.2% coverage is computed after excluding the uninspected upper lid of the T-pipe from the denominator, even though the preceding sentence says the upper lid remains uninspected and the "ideal feasible coverage" denominator is already intended to account for occlusions. This post-hoc exclusion inflates the reported number and makes the result difficult to interpret. The authors should report the raw coverage including the upper lid, state explicitly which surfaces are excluded from the denominator and why, and give per-object coverage numbers before averaging.
  3. [Section IV-B] The simulation comparison claim that the RL policy outperforms FUEL, GVI, and SWAP in inspection time rests on protocol details that are not fully controlled. The RL method receives a per-object inspection time limit and switches labels automatically, while the baselines are said to be fine-tuned to achieve greater than 95% coverage; it is unclear whether the plotted coverage for each baseline is measured at the same mission horizon or after a different termination criterion, and no per-run statistics or significance tests accompany the reported averages. The authors should specify the stopping rule for each method and report mean plus dispersion or per-run curves for all methods.
  4. [Tables I-III, Section IV-A] The ablation claims, such as the SVS map increasing coverage from 52.3% to 74.6% and GRU-512 increasing coverage from 59.8% to 74.6%, are presented as single point estimates without standard deviations, confidence intervals, or the number of independent training seeds. Given that the same statistic in Table I is reported across obstacle counts with substantial variation, adding dispersion and stating the number of seeds or checkpoints used is necessary to establish that the observed differences are not due to training variance.
minor comments (4)
  1. [Section III-C, Eq. (7)] The exact values of the reward parameters alpha, beta, gamma, delta, the desired inspection distance d_ref, and the collision distance d_coll used in the reported experiments are not given in the main text; please list them in a table or in the open-source documentation for reproducibility.
  2. [Section IV-A, Table I] The term "Timeout" is defined as the robot remaining collision-free until the end of the episode, which makes the near-100% timeout rates read as successful completion rather than failure; consider renaming this column to "No crash" or clarifying the definition.
  3. [Figure 5] Please specify whether the shaded region in Figure 5 represents variability across runs within an environment or across environments, and clarify what the two horizontal dashed lines denote for readers.
  4. [General] The notation q_t in Eq. (1) is described as a 4D orientation vector, but the quaternion convention and the relationship to the inertial frame I are not stated; a brief clarification would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the policy is evaluated against external baselines and real-world mesh reconstruction; self-citations are infrastructure only, and the reward/eval metric alignment is the task objective, not a fitted-input prediction.

full rationale

The paper's derivation chain is self-contained rather than circular. The RL policy is trained with reward R(st,at)=ft+vt+pt, where ft is the fraction of inspected semantic faces; the later coverage numbers measure the same facial-coverage quantity. This is the objective being learned rather than a disguised fit, and the paper additionally compares against independently tuned baselines (FUEL, GVI, SWAP) in simulation and against reconstructed real-world object meshes in deployment, so the central claims are externally grounded. Author-overlapping references (Aerial Gym Simulator [32] and prior NTNU platform papers) provide simulator and hardware infrastructure, and are not invoked as proof of the method's validity or uniqueness. The paper's 'first semantics-aware end-to-end RL inspection planning framework' claim is a novelty assertion, not an imported theorem, and no equation is defined in terms of the claimed result. The real-world use of HSV color ranges to produce segmentation masks is an engineering limitation of the semantic perception channel rather than a circular step; it weakens the 'previously unseen semantics' generalization claim but does not make the learned inspection behavior equivalent to its input by construction. The exclusion of the uninspected pipe lid from the feasible-coverage denominator is a measurement-choice concern, not a derivation-circularity concern. Overall, no step in the method reduces by definition or by self-citation to its own conclusions.

Assumptions & free parameters 7 free parameters · 5 assumptions · 0 invented entities

The central claim rests on a privileged face-mesh reward available only in simulation, on hand-tuned HSV color masks for real-world semantics, and on the assumption that policies trained on primitive shapes generalize to concave industrial objects. Several reward and safety parameters are chosen by hand. No new physical or algorithmic entities are introduced beyond the SVS map, which is a mapping structure rather than a postulated entity.

free parameters (7)
  • Reward parameter alpha (alpha)
    Eq. (7) face-mesh reward magnitude; a tuning parameter, value not reported.
  • Reward parameter beta (beta)
    Eq. (7) Gaussian sharpness around d_ref; a tuning parameter, value not reported.
  • Reward parameter gamma (gamma)
    Eq. (10) semantic-search reward scale; a tuning parameter, value not reported.
  • Reward parameter delta (delta)
    Eq. (10) visit-count decay; a tuning parameter, value not reported.
  • Desired inspection distance d_ref = 0.5-2.5 m training, 0.8-1.2 m evaluation
    Chosen hyperparameter defining inspection distance; appears in reward Eq. (7) and in the coverage evaluation metric.
  • Collision distance d_coll = 0.3 m
    Hand-chosen threshold for the collision penalty; related to robot size.
  • Maximum velocity and yaw-rate w_max = 1 m/s, 1 rad/s in simulation; 0.3 m/s, 0.6 rad/s in real experiments
    Action scaling in Eq. (5); chosen for platform safety and to reduce motion blur in real-world visual inspection.
assumptions (5)
  • ad hoc to paper The semantic object is a triangle mesh and a privileged per-face depth image is available during training to compute the face-mesh reward.
    Section III-C, Eq. (7)-(9): the reward requires the simulator to annotate each pixel with the face index, a signal not available to the robot at runtime.
  • ad hoc to paper Real-world 'semantics' can be isolated with fixed HSV color ranges.
    Section IV-C: masks are created from selected color ranges for the red pipe, blue barrel, and yellow hydrant; this assumes distinct colors and stable lighting.
  • domain assumption A policy trained on primitive shapes generalizes to unseen concave and industrial geometries.
    Section III-E trains on boxes, cylinders, spheres and one semantic with Nf=60; Section IV evaluates on bracket toes, pipes, and a spherical chamber. Demonstrated empirically but not guaranteed.
  • domain assumption The simulator dynamics, sensor noise, and disturbance model are faithful enough for sim2real transfer.
    Section III-E adds random wrenches, state noise, and depth-proportional image Gaussian noise, but faithfulness is assumed.
  • ad hoc to paper The 'ideal feasible coverage' denominator for each semantic can be computed correctly from occlusions and obstacles.
    Used in Sections IV-B and IV-C to normalize coverage; no algorithm or error analysis is provided, and the un-inspected upper lid was excluded from feasible surfaces after the fact.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Semantically-driven Deep Reinforcement Learning for Inspection Path Planning." pith.science (2026). https://pith.science/paper/RCGQUYJD

@misc{pith2026250514443,
  author       = {Pith},
  title        = {Pith review of: Semantically-driven Deep Reinforcement Learning for Inspection Path Planning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RCGQUYJD}},
  note         = {Machine review of arXiv:2505.14443}
}
read the original abstract

This paper introduces a novel semantics-aware inspection planning policy derived through deep reinforcement learning. Reflecting the fact that within autonomous informative path planning missions in unknown environments, it is often only a sparse set of objects of interest that need to be inspected, the method contributes an end-to-end policy that simultaneously performs semantic object visual inspection combined with collision-free navigation. Assuming access only to the instantaneous depth map, the associated segmentation image, the ego-centric local occupancy, and the history of past positions in the robot's neighborhood, the method demonstrates robust generalizability and successful crossing of the sim2real gap. Beyond simulations and extensive comparison studies, the approach is verified in experimental evaluations onboard a flying robot deployed in novel environments with previously unseen semantics and overall geometric configurations.

Figures

Figures reproduced from arXiv: 2505.14443 by the authors.

Figure 1
Figure 1. A flying robot conducting an inspection in an industrial environment. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. The proposed deep RL network for semantics-driven inspection. For a robot navigating an environment a target semantic object and obstacles, the [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Left: Inspection mission in the Aerial Gym Simulator. The environment consists of obstacles and a semantic of interest presented as a blue cuboid in this scenario. Right top-down: Images utilized during the training. the inspected object’s faces should be visually monitored. Additionally, we assume that a specific face of the semantic object has been seen if it falls within an image area, which we define as a focus … view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Left: WBT with three concave-shaped bracket toes. Right: Chemical plant with three objects: pipe, tank and exhaust. For each environment once the inspection time of the semantic is passed, the RL framework switches the semantic label to the next object of interest [PI…
Figure 5
Figure 5. Figure 5: The percentage of the cumulative surface of each semantic observed [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 7
Figure 7. Figure 7: Left: 3D perspective of the inspection mission with the path indicated by cyan arrows. Top row: Semantic depth masks of the objects as observed by the network and the corresponding RGB images from onboard camera captured during the mission. Bottom row: Instantaneous lo…
Figure 8
Figure 8. Figure 8: Surface coverage for each semantic is plotted against time, with dotted [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

36 extracted references · 34 canonical work pages

  1. [1]

    Topological path planning for autonomous information gathering,

    S. McCammon et al. , “Topological path planning for autonomous information gathering,” Autonomous Robots , vol. 45, no. 6, pp. 821– 842, 2021

    work page 2021

  2. [2]

    Hypergame-based adaptive behavior path planning for combined exploration and visual search,

    M. Dharmadhikari et al. , “Hypergame-based adaptive behavior path planning for combined exploration and visual search,” in 2021 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2021, pp. 269–275

    work page 2021

  3. [3]

    Fuel: Fast uav exploration using incremental frontier structure and hierarchical planning,

    B. Zhou et al. , “Fuel: Fast uav exploration using incremental frontier structure and hierarchical planning,” IEEE Robotics and Automation Letters, vol. 6, no. 2, pp. 779–786, 2021

    work page 2021

  4. [4]

    Tigris: An informed sampling-based algorithm for informative path planning,

    B. Moon et al. , “Tigris: An informed sampling-based algorithm for informative path planning,” in 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2022, pp. 5760–5766

    work page 2022

  5. [5]

    Autonomous exploration and general visual inspection of ship ballast water tanks using aerial robots,

    M. Dharmadhikari et al. , “Autonomous exploration and general visual inspection of ship ballast water tanks using aerial robots,” in 2023 21st International Conference on Advanced Robotics (ICAR) . IEEE, 2023, pp. 409–416

    work page 2023

  6. [6]

    A survey on coverage path planning for robotics,

    E. Galceran et al. , “A survey on coverage path planning for robotics,” Robotics and Autonomous systems, vol. 61, no. 12, pp. 1258–1276, 2013

    work page 2013

  7. [7]

    An efficient sampling-based method for online informative path planning in unknown environments,

    L. Schmid et al. , “An efficient sampling-based method for online informative path planning in unknown environments,” IEEE Robotics and Automation Letters , vol. 5, no. 2, pp. 1500–1507, 2020

    work page 2020

  8. [8]

    Semantic-aware informative path planning for efficient object search using mobile robot,

    C. Wang et al., “Semantic-aware informative path planning for efficient object search using mobile robot,” IEEE Transactions on Systems, Man, and Cybernetics: Systems , vol. 51, no. 8, pp. 5230–5243, 2019

    work page 2019

Show all 36 references
  1. [9]

    Semantics-aware exploration and inspection path planning,

    M. Dharmadhikari et al. , “Semantics-aware exploration and inspection path planning,” in 2023 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2023, pp. 3360–3367

    2023

  2. [10]

    Semantics-aware receding horizon planner for object- centric active mapping,

    L. Lu et al. , “Semantics-aware receding horizon planner for object- centric active mapping,” IEEE Robotics and Automation Letters , 2024

    2024

  3. [11]

    Zero-shot reinforcement learning on graphs for au- tonomous exploration under uncertainty,

    F. Chen et al. , “Zero-shot reinforcement learning on graphs for au- tonomous exploration under uncertainty,” in 2021 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2021, pp. 5193–5199

    2021

  4. [12]

    Learning active camera for multi-object navigation,

    P. Chen et al., “Learning active camera for multi-object navigation,” Ad- vances in Neural Information Processing Systems , vol. 35, pp. 28 670– 28 682, 2022

    2022

  5. [13]

    Where to look next: Learning viewpoint recom- mendations for informative trajectory planning,

    M. Lodel et al. , “Where to look next: Learning viewpoint recom- mendations for informative trajectory planning,” in 2022 International Conference on Robotics and Automation (ICRA) . IEEE, 2022, pp. 4466–4472

    2022

  6. [14]

    Catnipp: Context-aware attention-based network for informative path planning,

    Y . Cao et al. , “Catnipp: Context-aware attention-based network for informative path planning,” in Conference on Robot Learning . PMLR, 2023, pp. 1928–1937

    2023

  7. [15]

    Adaptive informative path planning with mul- timodal sensing,

    S. Choudhury et al. , “Adaptive informative path planning with mul- timodal sensing,” in Proceedings of the International Conference on Automated Planning and Scheduling , vol. 30, 2020, pp. 57–65

    2020

  8. [16]

    Active learning for uav-based semantic mapping,

    H. Blum et al., “Active learning for uav-based semantic mapping,” arXiv preprint arXiv:1908.11157, 2019

    1908 arXiv

  9. [17]

    A deep reinforcement learning approach with visual semantic navigation with memory for mobile robots in indoor home context,

    I. B. d. A. Santos et al. , “A deep reinforcement learning approach with visual semantic navigation with memory for mobile robots in indoor home context,” Journal of Intelligent & Robotic Systems , vol. 104, no. 3, p. 40, 2022

    2022

  10. [18]

    Enhancing navigational safety in crowded environ- ments using semantic-deep-reinforcement-learning-based navigation,

    L. K ¨astner et al. , “Enhancing navigational safety in crowded environ- ments using semantic-deep-reinforcement-learning-based navigation,” in 2022 IEEE International Symposium on Safety, Security, and Rescue Robotics (SSRR) . IEEE, 2022, pp. 87–93

    2022

  11. [19]

    Object goal navigation using goal-oriented seman- tic exploration,

    D. S. Chaplot et al., “Object goal navigation using goal-oriented seman- tic exploration,” Advances in Neural Information Processing Systems , vol. 33, pp. 4247–4258, 2020

    2020

  12. [20]

    Semantically-enhanced deep collision prediction for autonomous navigation using aerial robots,

    M. Kulkarni et al. , “Semantically-enhanced deep collision prediction for autonomous navigation using aerial robots,” in 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) . IEEE, 2023, pp. 3056–3063

    2023

  13. [21]

    Robotic learning for adaptive informative path planning,

    M. Popovic et al. , “Robotic learning for adaptive informative path planning,” arXiv preprint arXiv:2404.06940 , 2024

    2024 arXiv

  14. [22]

    Sc-explorer: Incremental 3d scene completion for safe and efficient exploration mapping and planning,

    L. Schmid et al. , “Sc-explorer: Incremental 3d scene completion for safe and efficient exploration mapping and planning,” arXiv preprint arXiv:2208.08307, 2022

    2022 arXiv

  15. [23]

    Learning exploration policies for navigation,

    T. Chen et al. , “Learning exploration policies for navigation,” arXiv preprint arXiv:1903.01959, 2019

    1903 arXiv

  16. [24]

    A data-driven model for interaction-aware pedestrian motion prediction in object cluttered environments,

    M. Pfeiffer et al., “A data-driven model for interaction-aware pedestrian motion prediction in object cluttered environments,” in 2018 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2018, pp. 5921–5928

    2018

  17. [25]

    Deep reinforcement learning robot for search and rescue applications: Exploration in unknown cluttered environments,

    F. Niroui et al. , “Deep reinforcement learning robot for search and rescue applications: Exploration in unknown cluttered environments,” IEEE Robotics and Automation Letters , vol. 4, no. 2, pp. 610–617, 2019

    2019

  18. [26]

    Deep reinforcement learning supervised autonomous ex- ploration in office environments,

    D. Zhu et al., “Deep reinforcement learning supervised autonomous ex- ploration in office environments,” in 2018 IEEE international conference on robotics and automation (ICRA) . IEEE, 2018, pp. 7548–7555

    2018

  19. [27]

    Receding horizon

    A. Bircher et al. , “Receding horizon” next-best-view” planner for 3d exploration,” in 2016 IEEE international conference on robotics and automation (ICRA) . IEEE, 2016, pp. 1462–1468

    2016

  20. [28]

    Receding horizon path planning for 3d exploration and surface inspection,

    ——, “Receding horizon path planning for 3d exploration and surface inspection,” Autonomous Robots , vol. 42, pp. 291–306, 2018

    2018

  21. [29]

    Sampling-based coverage path planning for inspection of complex structures,

    B. Englot et al., “Sampling-based coverage path planning for inspection of complex structures,” in Proceedings of the International Conference on Automated Planning and Scheduling , vol. 22, 2012, pp. 29–37

    2012

  22. [30]

    Sample factory: Egocentric 3d control from pixels at 100000 fps with asynchronous reinforcement learning,

    A. Petrenko et al., “Sample factory: Egocentric 3d control from pixels at 100000 fps with asynchronous reinforcement learning,” in International Conference on Machine Learning . PMLR, 2020, pp. 7652–7662

    2020

  23. [31]

    Impala: Scalable distributed deep-rl with importance weighted actor-learner architectures,

    L. Espeholt et al., “Impala: Scalable distributed deep-rl with importance weighted actor-learner architectures,” in International conference on machine learning . PMLR, 2018, pp. 1407–1416

    2018

  24. [32]

    Aerial gym simulator: A framework for highly parallelized simulation of aerial robots,

    M. Kulkarni, W. Rehberg, and K. Alexis, “Aerial gym simulator: A framework for highly parallelized simulation of aerial robots,” 2025

    2025

  25. [33]

    Control of complex maneuvers for a quadrotor uav using geometric methods on se (3),

    T. Lee et al., “Control of complex maneuvers for a quadrotor uav using geometric methods on se (3),” arXiv preprint arXiv:1003.2005 , 2010

    2005 arXiv

  26. [34]

    Design and use paradigms for gazebo, an open-source multi-robot simulator,

    N. Koenig et al., “Design and use paradigms for gazebo, an open-source multi-robot simulator,” in 2004 IEEE/RSJ international conference on intelligent robots and systems (IROS)(IEEE Cat. No. 04CH37566) , vol. 3. Ieee, 2004, pp. 2149–2154

    2004

  27. [35]

    V oxblox: Incremental 3d euclidean signed dis- tance fields for on-board mav planning,

    H. Oleynikova et al. , “V oxblox: Incremental 3d euclidean signed dis- tance fields for on-board mav planning,” in IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , 2017

    2017

  28. [36]

    Neural control barrier functions for safe navigation,

    M. Harms et al. , “Neural control barrier functions for safe navigation,” in 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2024, pp. 10 415–10 422

    2024

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.