Track A*: Fast Visibility-Aware Trajectory Planning for Active Target Tracking

Hanxuan Chen; Ji Pei; Kangli Wang

arxiv: 2605.05338 · v1 · submitted 2026-05-06 · 💻 cs.RO

Track A*: Fast Visibility-Aware Trajectory Planning for Active Target Tracking

Hanxuan Chen , Kangli Wang , Ji Pei This is my paper

Pith reviewed 2026-05-08 16:19 UTC · model grok-4.3

classification 💻 cs.RO

keywords visibility-aware planningactive target trackingtrajectory planningA* searchCARLA simulationbeam pruningoffline reference trajectoriesspatio-temporal grid

0 comments

The pith

Track A* finds high-quality visibility-aware target tracking trajectories over 20 times faster than standard A* while nearly preserving visibility scores.

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

The paper develops TA star as an offline planner that searches for trajectories maximizing visibility of a moving target while avoiding obstacles. It discretizes the problem into a four-dimensional grid of position and time then applies a layered DAG search with beam pruning, obstacle distance caching, and multi-ray visibility checks. On hundreds of CARLA scenarios the method reaches full convergence where a plain A* baseline fails in nearly half the cases, cuts average runtime by more than twenty times, and alters measured visibility by only a fraction of a percent on the cases both solve. This matters because repeatable, high-visibility reference trajectories are required both to build training datasets for trackers and to benchmark online planners under controlled conditions. The work explicitly accepts reduced theoretical optimality in exchange for the ability to finish large batches of scenarios in minutes rather than hours.

Core claim

TA star performs beam-pruned heuristic search over a layered directed acyclic graph that encodes a discretized 4D spatio-temporal state space. Three engineering additions—bounding-volume-hierarchy caching of obstacle distances, per-layer beam pruning, and a tunable multi-ray visibility evaluator—allow the planner to produce trajectories whose average visibility differs from an unoptimized A* baseline by only -0.15 percentage points on the 141 scenarios both methods solve, while reducing mean planning time by a factor of 23.0 and worst-case time by 11.8 and raising convergence from 56.9 % to 100 % across 248 controlled trials.

What carries the argument

Beam-pruned A* search on a layered DAG with BVH distance caching and configurable multi-ray visibility evaluation.

If this is right

Enables routine generation of large multi-modal tracking datasets from offline computation.
Supplies repeatable, high-convergence reference trajectories for benchmarking online planners.
Scales to 1000 scenarios across eight CARLA maps in under a minute of wall-clock time with modest parallelism.
Maintains visibility within 5 pp of baseline on every overlapping case while trading away strict optimality.
Reveals concrete failure modes in dense vegetation environments such as Town07.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same caching and pruning pattern could be reused for other 4D visibility or line-of-sight planning tasks.
Dynamic adjustment of beam width during search might further reduce runtime on easier scenarios.
Transfer to physical robots would require checking that discrete-grid visibility predictions match continuous sensor measurements.
The offline reference trajectories could serve as expert demonstrations for imitation learning of online trackers.

Load-bearing premise

The chosen grid resolution, beam-pruning thresholds, and multi-ray visibility evaluator produce trajectories whose real-world visibility and safety properties remain close to the discrete optimum.

What would settle it

Compute an exhaustive optimal visibility score on a small grid instance where the 4D state space can be searched completely, then compare that score against the visibility achieved by TA* on the same instance.

Figures

Figures reproduced from arXiv: 2605.05338 by Hanxuan Chen, Ji Pei, Kangli Wang.

**Figure 1.** Figure 1: Overview of the TA* algorithm pipeline (§III); the discretisation, view at source ↗

**Figure 2.** Figure 2: Per-map runtime and visibility distribution on cohort C1 (TA*-only view at source ↗

**Figure 3.** Figure 3: Cohort C2 (248-scenario controlled comparison) per-map runtime and view at source ↗

**Figure 6.** Figure 6: Frame-wise visibility on the worst-case scenario in cohort C3 (Town02 view at source ↗

**Figure 5.** Figure 5: Visibility change on cohort C3 (n = 141). The dashed reference line marks the 5 pp empirical envelope; no scenario crosses it. produces the largest visibility drop among baseline-converged scenarios (−4.32 pp); Town01, Town05, and Town06 are frame-wise identical on average. F. Optimization Ablation We additionally measured the contribution of the three engineering choices that distinguish TA* from the prio… view at source ↗

**Figure 7.** Figure 7: Single representative Town04 cohort-C2 run (path 0) under the matched view at source ↗

**Figure 8.** Figure 8: Optimization ablation: per-variant mean runtime (log scale, blue bars) view at source ↗

**Figure 9.** Figure 9: Speed-quality trade-off from the beam-width / ray-count sweep. The view at source ↗

read the original abstract

Offline reference trajectories for active target tracking are needed both for building multi-modal tracking datasets and for benchmarking online tracking planners under repeatable conditions. We present Track A star (TA star), an offline search-based trajectory planner that targets the visibility-aware target tracking objective on a discretized four-dimensional spatio-temporal grid (x, y, z, t). TA star combines a layered Directed Acyclic Graph (DAG) search with three engineering optimizations: cross-time obstacle distance caching against a Bounding Volume Hierarchy (BVH), per-layer beam pruning, and a configurable multi-ray visibility evaluator. TA star employs a beam-pruned heuristic search on this discrete graph to efficiently find high-quality tracking trajectories. While it trades strict theoretical optimality for practical scalability, our empirical results demonstrate robust, near-baseline visibility performance at a fraction of the computational cost. On a 1000-scenario stress test across eight CARLA Optimized maps, TA star converges on all scenarios and completes in 45 s using 32 workers; on a 248-scenario controlled comparison against an unoptimized priority-queue A star baseline (BinaryHeap implementation) under identical scenario inputs and a 5 x 10^6 expansion cap, TA star reduces mean planning time by 23.0x and worst-case planning time by 11.8x, while raising convergence from 56.9% to 100%. On the n=141 baseline-converged subset, TA star changes average visibility by only -0.15 percentage points (pp), with no scenario exceeding a 5 pp drop. We position TA star as a practical offline reference planner under these specific conditions, with limitations and failure cases discussed for environments such as Town07 dense vegetation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Track A* provides practical speedups for generating offline tracking trajectories but lacks sensitivity analysis on its key parameters.

read the letter

The main takeaway is that Track A* gives a workable way to generate visibility-aware reference trajectories offline, with a reported 23x speedup and near-identical visibility scores on CARLA data. What stands out is the engineering combination. They use a layered DAG to handle the time dimension in the 4D grid, cache distances with a BVH that works across time steps, apply beam pruning at each layer to cut down expansions, and evaluate visibility with multiple rays. This stack is tailored for the tracking problem and not something I have seen exactly in the prior work they cite. The results section does its job. The 248-scenario comparison shows clear gains in speed and convergence, and the larger 1000-scenario run confirms it scales to many maps. The visibility drop is small on average and bounded, which supports their claim of high-quality trajectories. They also discuss limitations in dense vegetation, which shows some awareness of where the approach might not transfer. Where it falls short is the lack of deeper validation on the design choices. The grid steps, beam width, and number of rays are set once and not varied, so we do not see how sensitive the visibility or time numbers are to those parameters. The stress-test concern about preservation under pruning holds up from what is shown; the paper does not provide that analysis. They mention issues in dense vegetation but do not quantify how the parameters affect outcomes there. Reproducibility is also limited without code or data. This leaves the central claim a bit narrower than the abstract suggests. This work is aimed at researchers who build multi-modal tracking datasets or need consistent benchmarks for online planners. Someone in that space could use the planner as a tool or the numbers as a reference point. I would recommend sending it for peer review. The empirical evidence is specific and the method is described clearly enough for others to build on or critique.

Referee Report

2 major / 2 minor

Summary. The manuscript presents Track A* (TA*), an offline search-based planner for visibility-aware active target tracking on a discretized 4D (x, y, z, t) spatio-temporal grid. It uses a layered DAG heuristic search augmented by three engineering optimizations—cross-time BVH caching for obstacle distances, per-layer beam pruning, and a configurable multi-ray visibility evaluator—while explicitly trading strict optimality for scalability. The central empirical claims are that TA* converges on all 1000+ CARLA scenarios (vs. 56.9% for baseline), reduces mean planning time by 23.0× and worst-case by 11.8× versus an unoptimized BinaryHeap A* under identical inputs and expansion caps, and changes average visibility by only −0.15 pp on the n=141 converged subset, with no scenario exceeding a 5 pp drop.

Significance. If the results hold, TA* supplies a practical offline reference generator for multi-modal tracking datasets and repeatable benchmarking of online planners. The scale of the evaluation (1000-scenario stress test across eight CARLA maps plus a 248-scenario controlled comparison) and the concrete timing/visibility metrics constitute a clear strength. The work is positioned as an engineering contribution that accepts discretization trade-offs, which is appropriate for the robotics application.

major comments (2)

[§5 / 248-scenario controlled comparison] §5 / 248-scenario controlled comparison: the 23.0× mean and 11.8× worst-case speed-ups and the −0.15 pp visibility change are reported only for one fixed choice of grid resolution, beam width, pruning threshold, and ray count, with no ablation isolating the contribution of BVH caching versus beam pruning versus the multi-ray evaluator and no sensitivity sweeps over these free parameters. Because the paper acknowledges that it trades strict optimality for scalability and flags failure modes in Town07 vegetation, the absence of such analysis leaves the claim that TA* trajectories remain close to the discrete optimum unsupported beyond the specific parameter set tested.
[§5 / visibility metric definition] §5 / visibility metric definition: the multi-ray visibility evaluator is described as “configurable,” yet no quantitative sensitivity of the reported visibility scores to ray count or ray placement is provided. If visibility is sensitive to these choices (as the skeptic note on Town07 vegetation suggests), the −0.15 pp average change cannot be taken as evidence that the pruned trajectories preserve the visibility properties of the underlying discrete optimum.

minor comments (2)

[Abstract] Abstract: mean timing and visibility figures are given without error bars, standard deviations, or any indication of statistical significance, which weakens the reader’s ability to judge the reliability of the 23.0× and −0.15 pp claims.
[Methods] Methods: the precise definition of the layered DAG, the beam-pruning rule, and the cross-time BVH caching data structure would benefit from a small pseudocode listing or diagram to make the three optimizations reproducible from the text alone.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback and positive overall assessment of the work as a practical engineering contribution. We agree that the absence of ablations and sensitivity analysis limits the strength of claims about robustness beyond the tested configuration. We address each major comment below and will incorporate the requested analyses in the revised manuscript.

read point-by-point responses

Referee: [§5 / 248-scenario controlled comparison] §5 / 248-scenario controlled comparison: the 23.0× mean and 11.8× worst-case speed-ups and the −0.15 pp visibility change are reported only for one fixed choice of grid resolution, beam width, pruning threshold, and ray count, with no ablation isolating the contribution of BVH caching versus beam pruning versus the multi-ray evaluator and no sensitivity sweeps over these free parameters. Because the paper acknowledges that it trades strict optimality for scalability and flags failure modes in Town07 vegetation, the absence of such analysis leaves the claim that TA* trajectories remain close to the discrete optimum unsupported beyond the specific parameter set tested.

Authors: We acknowledge that the 248-scenario results reflect the combined optimizations under one fixed parameter set chosen via preliminary tuning for practical performance. The controlled comparison (identical inputs, grid, expansion cap, and visibility evaluator) directly measures the net effect on time and visibility relative to unoptimized A*. We agree, however, that isolating individual contributions and providing sensitivity sweeps would better substantiate robustness of the near-baseline visibility claim. In the revision we will add an ablation study and parameter sensitivity analysis (beam width, pruning threshold, ray count) on a representative subset of scenarios, reporting effects on planning time and visibility. revision: yes
Referee: [§5 / visibility metric definition] §5 / visibility metric definition: the multi-ray visibility evaluator is described as “configurable,” yet no quantitative sensitivity of the reported visibility scores to ray count or ray placement is provided. If visibility is sensitive to these choices (as the skeptic note on Town07 vegetation suggests), the −0.15 pp average change cannot be taken as evidence that the pruned trajectories preserve the visibility properties of the underlying discrete optimum.

Authors: The multi-ray evaluator approximates continuous visibility via a fixed sampling pattern chosen for consistency across all experiments. The reported −0.15 pp average change is measured under this identical approximation for both TA* and the baseline, so the relative comparison remains valid. We concur that quantitative sensitivity to ray count and placement would strengthen evidence that pruning preserves visibility properties. We will add a sensitivity study in the revised manuscript showing visibility score variation with different ray counts and placements on the same scenarios. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical benchmark of heuristic planner with explicit optimality trade-off

full rationale

The paper introduces TA* as a beam-pruned DAG search on a 4D grid with BVH caching and multi-ray visibility, explicitly framed as a practical heuristic that trades strict optimality for scalability. All reported claims (23.0x mean speedup, 100% convergence, -0.15 pp visibility delta) are direct empirical measurements on CARLA scenarios against an unoptimized BinaryHeap A* baseline under identical inputs and expansion caps. No fitted parameters are renamed as predictions, no self-citations bear the central result, and the algorithm derivation does not reduce to its own outputs by construction. The evaluation is self-contained against external simulation benchmarks.

Axiom & Free-Parameter Ledger

3 free parameters · 2 axioms · 0 invented entities

The planner rests on standard A* graph-search assumptions plus three engineering choices whose correctness is validated only empirically: the 4D grid discretization is fine enough, beam pruning does not eliminate all near-optimal paths, and the multi-ray visibility test approximates real sensor visibility.

free parameters (3)

grid resolution (x,y,z,t)
Chosen discretization step sizes that define the search graph size and approximation quality.
beam width / pruning threshold
Configurable parameter that trades completeness for speed; directly affects which trajectories are considered.
number of visibility rays
Hyper-parameter controlling the fidelity of the visibility evaluator.

axioms (2)

standard math A* with admissible heuristic finds optimal paths on the discrete graph
Invoked implicitly when claiming the baseline is optimal on the same graph.
domain assumption Beam pruning preserves near-optimal solutions for the visibility objective
Stated as a practical trade-off; no formal proof is given.

pith-pipeline@v0.9.0 · 5609 in / 1537 out tokens · 28665 ms · 2026-05-08T16:19:33.074349+00:00 · methodology

discussion (0)

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

CosFly: Plan in the Matrix, Fly in the World
cs.RO 2026-05 unverdicted novelty 6.0

CosFly introduces a box-structured planning and multimodal simulation pipeline for aerial target tracking in CARLA, paired with the public CosFly-Track dataset containing 250 trajectories and approximately 100,000 ren...

Reference graph

Works this paper leans on

16 extracted references · 16 canonical work pages · cited by 1 Pith paper

[1]

Towards a robust aerial cinematography platform: Localizing and tracking moving targets in unstructured environments,

R. Bonatti, W. Wang, C. Ho, A. Ahuja, M. Gschwindt, E. Camci, E. Kayacan, S. Choudhury, and S. Scherer, “Towards a robust aerial cinematography platform: Localizing and tracking moving targets in unstructured environments,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2019, pp. 229–236

work page 2019
[2]

Visibility-aware trajectory optimization with application to aerial tracking,

Q. Wang, Y . Chen, and F. Gao, “Visibility-aware trajectory optimization with application to aerial tracking,” inIEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS). IEEE, 2021, pp. 5549– 5556

work page 2021
[3]

Real-time obstacle avoidance for manipulators and mobile robots,

O. Khatib, “Real-time obstacle avoidance for manipulators and mobile robots,”The International Journal of Robotics Research, vol. 5, no. 1, pp. 90–98, 1986

work page 1986
[4]

Predictive active steering control for autonomous vehicle systems,

P. Falcone, F. Borrelli, J. Asgari, H. E. Tseng, and D. Hrovat, “Predictive active steering control for autonomous vehicle systems,”IEEE Trans- actions on Control Systems Technology, vol. 15, no. 3, pp. 566–580, 2007

work page 2007
[5]

A formal basis for the heuristic determination of minimum cost paths,

P. E. Hart, N. J. Nilsson, and B. Raphael, “A formal basis for the heuristic determination of minimum cost paths,”IEEE Transactions on Systems Science and Cybernetics, vol. 4, no. 2, pp. 100–107, 1968

work page 1968
[6]

Sampling-based algorithms for optimal motion planning,

S. Karaman and E. Frazzoli, “Sampling-based algorithms for optimal motion planning,”The International Journal of Robotics Research, vol. 30, no. 7, pp. 846–894, 2011

work page 2011
[7]

Enable faster and smoother spatio-temporal trajectory planning for autonomous vehicles in constrained dynamic environment,

L. Xin, P. Wang, C.-Y . Chan, J. Chen, S. E. Li, and B. Jiang, “Enable faster and smoother spatio-temporal trajectory planning for autonomous vehicles in constrained dynamic environment,”Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering, vol. 235, no. 10-11, pp. 2838–2853, 2021

work page 2021
[8]

Efficient spatiotemporal graph search for local trajectory planning on oval race tracks,

M. Rowold and J. Betz, “Efficient spatiotemporal graph search for local trajectory planning on oval race tracks,”Actuators, vol. 11, no. 11, p. 319, 2022

work page 2022
[9]

Spatio-temporal trajectory planning using search and optimizing method for autonomous driving,

L. Zhong, Z. Li, Z. Wang, Y . Zhenget al., “Spatio-temporal trajectory planning using search and optimizing method for autonomous driving,” inSAE Technical Paper. SAE International, 2024

work page 2024
[10]

Eva-planner: Environmental adaptive quadrotor planning,

L. Wang, H. Wei, T. Yang, Y . Tang, L. Yu, and X. Yang, “Eva-planner: Environmental adaptive quadrotor planning,”IEEE Transactions on Intelligent Vehicles, 2023, cited as “Eva-Tracker” in the related work section

work page 2023
[11]

Svpto: safe visibility-guided perception-aware trajectory optimization for aerial tracking,

H. Wang, Y . Zhang, L. Hanet al., “Svpto: safe visibility-guided perception-aware trajectory optimization for aerial tracking,”IEEE Transactions on Industrial Informatics, 2023

work page 2023
[12]

Path planning for uav tracking target based on improved a-star algorithm,

Y . Cai, Q. Xi, X. Xing, H. Gui, and Q. Liu, “Path planning for uav tracking target based on improved a-star algorithm,” inIEEE International Conference on Unmanned Systems (ICUS). IEEE, 2019, pp. 1–5

work page 2019
[13]

Uav target tracking: a survey,

P. Wu, Y . Li, and D. Xue, “Uav target tracking: a survey,”Artificial Intelligence Review, vol. 58, no. 1, pp. 1–45, 2025

work page 2025
[14]

Theta*: Any-angle path planning on grids,

A. Nash, K. Daniel, S. Koenig, and A. Felner, “Theta*: Any-angle path planning on grids,” inAAAI Conference on Artificial Intelligence, 2007, pp. 1177–1183

work page 2007
[15]

D* lite,

S. Koenig and M. Likhachev, “D* lite,” inAAAI Conference on Artificial Intelligence, 2002

work page 2002
[16]

CARLA: An open urban driving simulator,

A. Dosovitskiy, G. Ros, F. Codevilla, A. Lopez, and V . Koltun, “CARLA: An open urban driving simulator,” inProceedings of the 1st Annual Conference on Robot Learning, 2017, pp. 1–16

work page 2017

[1] [1]

Towards a robust aerial cinematography platform: Localizing and tracking moving targets in unstructured environments,

R. Bonatti, W. Wang, C. Ho, A. Ahuja, M. Gschwindt, E. Camci, E. Kayacan, S. Choudhury, and S. Scherer, “Towards a robust aerial cinematography platform: Localizing and tracking moving targets in unstructured environments,” inIEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2019, pp. 229–236

work page 2019

[2] [2]

Visibility-aware trajectory optimization with application to aerial tracking,

Q. Wang, Y . Chen, and F. Gao, “Visibility-aware trajectory optimization with application to aerial tracking,” inIEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS). IEEE, 2021, pp. 5549– 5556

work page 2021

[3] [3]

Real-time obstacle avoidance for manipulators and mobile robots,

O. Khatib, “Real-time obstacle avoidance for manipulators and mobile robots,”The International Journal of Robotics Research, vol. 5, no. 1, pp. 90–98, 1986

work page 1986

[4] [4]

Predictive active steering control for autonomous vehicle systems,

P. Falcone, F. Borrelli, J. Asgari, H. E. Tseng, and D. Hrovat, “Predictive active steering control for autonomous vehicle systems,”IEEE Trans- actions on Control Systems Technology, vol. 15, no. 3, pp. 566–580, 2007

work page 2007

[5] [5]

A formal basis for the heuristic determination of minimum cost paths,

P. E. Hart, N. J. Nilsson, and B. Raphael, “A formal basis for the heuristic determination of minimum cost paths,”IEEE Transactions on Systems Science and Cybernetics, vol. 4, no. 2, pp. 100–107, 1968

work page 1968

[6] [6]

Sampling-based algorithms for optimal motion planning,

S. Karaman and E. Frazzoli, “Sampling-based algorithms for optimal motion planning,”The International Journal of Robotics Research, vol. 30, no. 7, pp. 846–894, 2011

work page 2011

[7] [7]

Enable faster and smoother spatio-temporal trajectory planning for autonomous vehicles in constrained dynamic environment,

L. Xin, P. Wang, C.-Y . Chan, J. Chen, S. E. Li, and B. Jiang, “Enable faster and smoother spatio-temporal trajectory planning for autonomous vehicles in constrained dynamic environment,”Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering, vol. 235, no. 10-11, pp. 2838–2853, 2021

work page 2021

[8] [8]

Efficient spatiotemporal graph search for local trajectory planning on oval race tracks,

M. Rowold and J. Betz, “Efficient spatiotemporal graph search for local trajectory planning on oval race tracks,”Actuators, vol. 11, no. 11, p. 319, 2022

work page 2022

[9] [9]

Spatio-temporal trajectory planning using search and optimizing method for autonomous driving,

L. Zhong, Z. Li, Z. Wang, Y . Zhenget al., “Spatio-temporal trajectory planning using search and optimizing method for autonomous driving,” inSAE Technical Paper. SAE International, 2024

work page 2024

[10] [10]

Eva-planner: Environmental adaptive quadrotor planning,

L. Wang, H. Wei, T. Yang, Y . Tang, L. Yu, and X. Yang, “Eva-planner: Environmental adaptive quadrotor planning,”IEEE Transactions on Intelligent Vehicles, 2023, cited as “Eva-Tracker” in the related work section

work page 2023

[11] [11]

Svpto: safe visibility-guided perception-aware trajectory optimization for aerial tracking,

H. Wang, Y . Zhang, L. Hanet al., “Svpto: safe visibility-guided perception-aware trajectory optimization for aerial tracking,”IEEE Transactions on Industrial Informatics, 2023

work page 2023

[12] [12]

Path planning for uav tracking target based on improved a-star algorithm,

Y . Cai, Q. Xi, X. Xing, H. Gui, and Q. Liu, “Path planning for uav tracking target based on improved a-star algorithm,” inIEEE International Conference on Unmanned Systems (ICUS). IEEE, 2019, pp. 1–5

work page 2019

[13] [13]

Uav target tracking: a survey,

P. Wu, Y . Li, and D. Xue, “Uav target tracking: a survey,”Artificial Intelligence Review, vol. 58, no. 1, pp. 1–45, 2025

work page 2025

[14] [14]

Theta*: Any-angle path planning on grids,

A. Nash, K. Daniel, S. Koenig, and A. Felner, “Theta*: Any-angle path planning on grids,” inAAAI Conference on Artificial Intelligence, 2007, pp. 1177–1183

work page 2007

[15] [15]

D* lite,

S. Koenig and M. Likhachev, “D* lite,” inAAAI Conference on Artificial Intelligence, 2002

work page 2002

[16] [16]

CARLA: An open urban driving simulator,

A. Dosovitskiy, G. Ros, F. Codevilla, A. Lopez, and V . Koltun, “CARLA: An open urban driving simulator,” inProceedings of the 1st Annual Conference on Robot Learning, 2017, pp. 1–16

work page 2017