Pith

open record

sign in

arxiv: 2607.06964 · v1 · pith:56TFP73F · submitted 2026-07-08 · cs.RO · cs.AI

End-to-End LLM Flight Planning with RAG-based Memory and Multi-modal Coach Agent

Amin Tabrizian , Arsyi Aziz , Aarifah Ullah , Mahyar Ghazanfari , Pouria Razzaghi , Peng Wei This is my paper

Reviewed by Pith T0 review T1 audit T2 compute T3 formal T4 kernel 2026-07-09 00:52 UTCglm-5.2pith:56TFP73Frecord.jsonopen to challenge →

classification cs.RO cs.AI
keywords LLM planningflight planningeVTOLretrieval-augmented generationmulti-modal agentpreference alignmentUAVadvanced air mobility
0
0 comments X

The pith

LLM flight planner with memory and vision coach hits 94% validity

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

The paper presents FRAMe, a flight-planning system for eVTOL aircraft built around a large language model that translates natural-language operator preferences into sequences of geographic waypoints. The system pairs the planner LLM with two add-on modules: a retrieval-augmented memory that supplies relevant past flight plans as in-context examples, and a multi-modal coach agent that uses rule-based geometry checks to verify plan validity and a vision-capable LLM to assess whether the route aligns with the operator's stated preference. The central claim is that each component contributes non-redundantly: across four LLMs and three difficulty levels, the full system (RAG plus coach) yields the highest plan validity for every planner tested, reaching up to 93.8% aggregate and 99% on easy scenarios for the strongest planner. The coach is shown to be load-bearing rather than decorative—for one model, retrieval alone slightly regressed validity and the coach recovered the gain. The paper also argues that LLM-based planners are preference-aware where classical algorithms like A* are not: A* achieves perfect validity but is blind to operator intent, producing routes with 3–5× more waypoints and far less clearance from hazards than even the baseline LLM configurations. Preference capture is demonstrated on three measurable objectives—minimize distance, minimize waypoints, maximize clearance—with gains that are model-specific and constrained by how much headroom each planner has at baseline.

Core claim

The paper's central result is that combining retrieval-augmented memory with a multi-modal coach agent makes LLM-generated flight plans both more valid and more preference-aligned than either component alone, and that this combination is necessary because retrieval without validation can actively harm validity for certain models. The coach agent's three-stage pipeline—geometric validity checks, vision-based preference assessment, and optional human feedback—creates a feedback loop where evaluated plans are stored in the memory database, progressively improving future retrievals. The ablation across four LLMs isolates each module's contribution: RAG helps most when the planner already follows

What carries the argument

three-stage coach agent (geometric validity + vision-based preference alignment + operator feedback) combined with cosine-similarity RAG retrieval over prior plans for geometrically identical scenarios

If this is right

  • The coach-as-validator pattern could generalize to other LLM planning domains where outputs must satisfy hard geometric or physical constraints—robotics, logistics routing, autonomous driving—provided that rule-based validity checks exist alongside the LLM judge.
  • The finding that retrieval alone can regress performance without a validation gate suggests that RAG-based planning systems need a filtering or verification step between retrieval and use, particularly for weaker or less instruction-following models.
  • The modular ablation protocol (warmup-seeded database, then read-only sweep) offers a reusable evaluation methodology for separating the effects of memory, verification, and base model capability in any system that combines LLM planning with retrieval.
  • The model-specific difficulty profiles (e.g., one model peaking on hard scenarios while another degrades) suggest that planner-model selection for safety-critical applications should be conditioned on the operational complexity envelope, not just aggregate benchmarks.

Where Pith is reading between the lines

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

  • If the coach's vision-based preference alignment is biased by the judge LLM's priors—as the paper acknowledges—then the preference-capture results may partly reflect the judge's expectations rather than purely objective alignment, which would mean the reported gains are an upper bound on true preference capture.
  • The system's reliance on exact scenario matching for RAG retrieval (same flyzone, polygons, origin, destination) means the memory only helps when the same geographic problem recurs with different preferences; this limits generalization to novel airspace configurations where no prior plans exist.
  • The absence of dynamic constraints (traffic, energy limits, vehicle dynamics) means the validity metric measures geometric feasibility only; plans that pass the coach's checks could still be infeasible for real eVTOL operations, so the 93.8% validity rate should not be read as an operational readiness figure.
  • The monotonic decline in clearance for one model under augmentation suggests a potential tension between the coach's validity-first corrections and preference optimization that could become more pronounced as the number of simultaneous preferences grows.

Load-bearing premise

The preference-alignment assessment relies on a vision-capable LLM judging the planner's output, which inherits the judge model's hidden biases—including a tendency to favor outputs resembling its own style—with no human validation of whether the coach's alignment verdicts actually match what human operators would prefer.

What would settle it

If the coach agent's vision-based preference judgments were shown by human raters to disagree systematically with human operator preferences on a representative subset of plans, the preference-capture claims would be undermined regardless of the geometric validity results.

Figures

Figures reproduced from arXiv: 2607.06964 by Aarifah Ullah, Amin Tabrizian, Arsyi Aziz, Mahyar Ghazanfari, Peng Wei, Pouria Razzaghi.

Figure 1
Figure 1. Figure 1: FRAMe System Architecture: The planner LLM generates a flight plan from a natural language prompt and scenario data. A RAG module provides relevant context to the LLM, such as previous similar planning experiences and their evaluations. The multi-modal LLM coach agent evaluates the planner’s proposed route for safety and compliance with the preference. Then, it supplies review that will be combined with th… view at source ↗
Figure 2
Figure 2. Figure 2: A sample of the generated image of a flight plan for the multi-modal coach agent to review. 2.3. Retrieval Augmented Generation To augment the prompt with proper context about the current flight planning problem, we implement the RAG module. When enabled, the system queries a vector database to re￾trieve past flight plans that most closely resemble the current setting and the operator’s stated preferences.… view at source ↗
Figure 3
Figure 3. Figure 3: Overview tab in the web user interface: a leaflet map shows the computed route (black) with origin (green), destination (blue), and no-fly zone boundaries (red). AI-reasoning will be provided in this tab. 3.1. Experimental Setup We evaluate FRAMe on flight scenarios in the Dallas–Fort Worth metropolitan area grouped into three difficulty levels— Easy, Medium, and Hard—that differ in the number and geometry… view at source ↗
Figure 4
Figure 4. Figure 4: Validity rate (%) by condition for all planners. Conditions are Baseline, +RAG, and +RAG+Coach. Baseline RAG RAG+Coach 190 191 192 193 194 195 Distance (km) 192.9 193.5 193.3 194.4 193.8 195.0 190.5 190.1 190.0 192.8 194.1 193.4 Preference: distance (target ) o3-mini o4-mini gpt-5.4 deepseek-r1 Baseline RAG RAG+Coach 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4.0 Waypoints 4.1 3.8 3.6 3.5 3.5 3.5 4.0 3.8 3.5 3.3 3.4 3.3 … view at source ↗
Figure 5
Figure 5. Figure 5: Mean of the preference-relevant metric versus condition for each of the three preferences (valid plans only). Desired direction marked above each panel. achieves the highest validity. The three reasoning mod￾els converge to 88.9–93.8% under the full system (o3-mini: 79.6%→79.1%→88.9%, o4-mini: 84.4%→88.4%→90.7%, DeepSeek-R1: 83.6%→87.1%→93.8%), while GPT-5.4 lags considerably, improving from a markedly low… view at source ↗
Figure 6
Figure 6. Figure 6: Easy scenario, clearance preference. The generated image of the flight plan. C.2. Medium scenario, minimize distance preference FRAMe Input (SYS + USER + OPERATOR) SYS: You are a flight planner for an eVTOL aircraft. The user will provide you with hazardous polygon information and request a flight plan from an origin to a destination. You must generate a flight plan as a list of waypoints starting from the… view at source ↗
Figure 7
Figure 7. Figure 7: Medium scenario, minimize distance preference. The generated image of the flight plan. 14 [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Hard scenario, minimize waypoints preference. The generated image of the flight plan. o3-mini o4-mini gpt-5.4 deepseek-r1 0 20 40 60 80 100 Validity % 93 89 60 99 92 87 52 97 81 96 55 85 Easy Medium Hard [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Plan validity (%) of the full FRAMe system (+RAG+Coach) broken down by difficulty level (Easy, Medium, Hard) for each planner. The relationship between difficulty and validity is model-specific: DeepSeek-R1 peaks on Easy/Medium (98.7%/97.3%) while o4-mini peaks on Hard (96.0%), and GPT-5.4 remains consistently the weakest planner across all difficulty levels (52–60%). becomes less reliable when navigating … view at source ↗
Figure 10
Figure 10. Figure 10: Mean metric values by preference and condition for each model. Each cell reports the average over valid plans (o3-mini and o4-mini). 17 [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Mean metric values by preference and condition for each model. Each cell reports the average over valid plans(gpt-5.4 and deepseek-r1). 18 [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
read the original abstract

Bridging the gap between human pilot intent and autonomous flight operation is critical for real-world electric vertical takeoff and landing (eVTOL) aircraft deployment. Flight planning traditionally relies on classic algorithms that struggle to incorporate flexible human preferences. We present FRAMe, an End-to-End Large Language Model (LLM) Flight Planning tool with RAG-based Memory and Multi-modal Coach Agent. Our system integrates a planner LLM with a multi-modal coach agent and retrieval augmented generation (RAG)-based memory to generate flight plans that satisfy mission constraints while aligning with human flight operator preferences. We demonstrate the system in a range of real-world-inspired scenarios of varying difficulty levels. Across four LLMs, the full FRAMe system (RAG and coach) yields the highest validity for every planner (up to 93.8% aggregate, 99% on Easy scenarios for the strongest planner) and shifts preference-relevant metrics in the operator-favored direction where the metric has headroom. FRAMe signifies how advanced LLMs can be deployed for human-centric mission planning, translating natural language instructions into safe, efficient, and flexible flight routes. The code is available at: github.com/amin-tabrizian/FlightPlanningLLMs

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.

Referee Report

1 major / 8 minor

Summary. The paper presents FRAMe, an end-to-end LLM-based flight planning system for eVTOL aircraft that integrates a planner LLM with a RAG-based memory module and a multi-modal coach agent. The coach agent performs geometric validity checks and preference-alignment assessment using vision capabilities. The system is evaluated across four LLMs (o3-mini, o4-mini, GPT-5.4, DeepSeek-R1), three difficulty levels, and four conditions (A*, Baseline, +RAG, +RAG+Coach) using a warmup/ablation protocol. The central claims are: (1) the full system yields the highest validity for every planner tested, and (2) the full system shifts preference-relevant metrics in the operator-favored direction where headroom exists. Evaluation uses geometrically computed metrics (distance, waypoint count, polygon clearance) rather than human annotation.

Significance. The paper makes a solid contribution to LLM-based planning in the aviation domain. Strengths include: (1) a systematic ablation protocol with a warmup phase that seeds the RAG database followed by a read-only evaluation phase, cleanly isolating RAG and coach contributions; (2) an annotator-free preference evaluation framework using three geometrically computed metrics, avoiding LLM-as-judge for the reported preference-capture results; (3) evaluation across four state-of-the-art LLMs and three difficulty levels with a classical A* baseline for comparison; (4) reproducible code released at a public repository. The finding that the coach is load-bearing on top of RAG (particularly for o3-mini, where RAG alone regresses validity) is a useful empirical result for the community.

major comments (1)
  1. Abstract and Section 3.2: The central preference-capture claim states the full system 'shifts preference-relevant metrics in the operator-favored direction where the metric has headroom.' However, the paper's own data in Figure 5 shows that o3-mini under the clearance preference exhibits a monotonic decline from 5.88 km (Baseline) to 5.35 km (+RAG) to 4.52 km (+RAG+Coach)—a 23% decrease in the wrong direction. A* achieves 1.91 km, so 5.88 km is far from any ceiling, meaning headroom clearly exists. The paper acknowledges this decline but frames it as the coach 'prioritizing correcting validity violations over maximizing polygon separation.' This is an admission that the system fails to capture the stated preference for this model. The abstract's unconditional claim ('shifts preference-relevant metrics in the operator-favored direction where the metric has headroom') is contradicted by o3
minor comments (8)
  1. Section 2.3: The RAG retrieval restricts candidates to prior records with identical scenario geometry. This is a strong restriction that limits the generality of the memory system. Please discuss whether this design choice limits applicability to novel scenarios and acknowledge it as a limitation, perhaps in Section 4.
  2. Section 3.1: The number of scenarios per difficulty level and the total number of evaluation runs per condition are not clearly stated. Please specify the exact scenario counts and the number of runs per cell in the evaluation grid.
  3. Algorithm 1, line 9: The notation EmbedAll(cands) suggests embedding all candidate preference texts, but the text in Section 2.3 describes embedding only the operator preference. Please clarify the relationship.
  4. Section 2.2: The coach agent uses o4-mini for preference alignment, but it is unclear whether the same o4-mini model is also used as a planner in the experiments. If so, this creates a self-evaluation dynamic that should be discussed, at least in the limitations.
  5. Figure 5: The y-axis ranges are narrow, which visually exaggerates small differences. Consider widening the ranges or annotating the percent change from Baseline.
  6. Appendix D, Figure 9: The per-difficulty breakdown is informative. Consider moving at least a summary of this analysis to the main text, as it shows where the coach and RAG contribute most.
  7. References: The citation for 'ichter et al., 2023' (SayCan) appears to have a lowercase initial. Please verify.
  8. Section 1: The term 'GPT-5.4' is used; please verify this is the correct model name and cite it more precisely.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and agree the abstract overstates the preference-capture claim. We will revise.

read point-by-point responses

  1. Referee: The abstract's unconditional claim that the full system 'shifts preference-relevant metrics in the operator-favored direction where the metric has headroom' is contradicted by o3-mini under the clearance preference, which declines monotonically from 5.88 km (Baseline) to 4.52 km (+RAG+Coach)—a 23% decrease in the wrong direction, with clear headroom (A* achieves 1.91 km). The paper acknowledges this but the abstract does not qualify the claim.

    Authors: The referee is correct. The abstract's claim is stated unconditionally and is directly contradicted by the o3-mini clearance data in Figure 5. We acknowledge this as a genuine overstatement in the abstract that we will correct. Specifically, we will revise the abstract to read that the full system 'shifts preference-relevant metrics in the operator-favored direction where the metric has headroom for most model–preference combinations,' and will add an explicit caveat noting the o3-mini clearance exception. The body text (Section 3.2) already discusses this case honestly, noting that 'coach-driven revisions appear to prioritize correcting validity violations over maximizing polygon separation,' but the abstract must reflect this qualification as well. We agree this is not merely a rounding or framing issue: o3-mini's clearance under the clearance preference moves 23% in the wrong direction with substantial headroom, and the abstract should not claim otherwise. revision: yes

Circularity Check

0 steps flagged

No significant circularity: the paper's central claims are evaluated against geometrically computed metrics and an external A* baseline, with self-citations used only for prior prompting strategies.

full rationale

The paper's main results—plan validity rates and preference-metric shifts—are computed using rule-based geometric tools (polygon intersection, flyzone containment, distance/clearance calculations) that are independent of the LLM components being evaluated. The preference-capture metrics in Figure 5 and Table 1 are computed geometrically, not by the LLM-as-judge, so the evaluation does not reduce to the system's own outputs. The warmup/ablation protocol (Section 3.1) uses a read-only database during ablation, which cleanly isolates RAG and coach contributions without feeding results back into the evaluation. The self-citations to Tabrizian et al. (2024, 2025) are used to reference prior work on LLM flight planning and prompting strategies (specifically, the Zero-shot CoT prompt in Appendix A), but these are architectural choices, not load-bearing claims that the present paper's results depend on for validity. The o4-mini vision agent operates within the system pipeline (coach feedback stored in RAG), but the reported preference-capture results do not depend on its verdicts. The one minor concern is that the coach agent (o4-mini) both evaluates plans and contributes to the RAG database that the planner draws from, creating a potential feedback loop, but this is an architectural design choice rather than a circular derivation: the evaluation metrics are computed independently of the coach's alignment verdicts. No step in the paper's derivation chain reduces to its own inputs by construction. The paper is self-contained against external benchmarks (A* baseline, geometric validity checker) and the self-citations are not load-bearing for the central claims. Score 1 reflects the minor self-citation for prompting strategy, which does not affect the evaluation's independence.

Axiom & Free-Parameter Ledger

2 free parameters · 4 axioms · 0 invented entities

The system has two manually set parameters (K=2, 150 warmup runs) that are not derived from optimization. The axioms are domain assumptions about LLM parseability, geometric validity sufficiency, and LLM-as-judge reliability. No new physical entities, particles, or forces are introduced. The system is an engineering architecture, not a theoretical derivation.

free parameters (2)
  • K (retrieval count) = 2
    The number of prior flight plans retrieved by the RAG module is set to K=2 (Section 3.1). This is a design choice not derived from first principles or optimized via hyperparameter search.
  • Warmup runs = 150
    The warmup phase runs 150 plans to seed the RAG database (Section 3.1). This is a fixed quantity chosen by the authors.
axioms (4)
  • domain assumption LLM outputs can be parsed into structured waypoint sequences
    The system assumes the planner LLM's response can be reliably parsed into geographic coordinates (Section 2.1, Algorithm 1 line 16). Parse failures are not discussed as a failure mode.
  • domain assumption Geometric validity checks are sufficient for flight safety
    Validity is defined as waypoints within flyzone, correct origin/destination, and no segment through no-fly zones (Section 2.2). The paper acknowledges in Section 4 that dynamic traffic, energy limits, and vehicle dynamics are not modeled.
  • domain assumption LLM-as-judge provides reliable preference alignment assessment
    The coach's vision-based alignment verdict is used as ground truth for preference capture (Section 2.2, Section 4). The paper acknowledges this inherits biases but does not validate against human raters.
  • standard math Cosine similarity over preference embeddings ranks relevance within identical scenario geometry
    The RAG module uses cosine distance over preference text embeddings to rank prior plans for the same scenario (Section 2.3). This is a standard retrieval assumption.

pith-pipeline@v1.1.0-glm · 22098 in / 2482 out tokens · 512624 ms · 2026-07-09T00:52:55.257939+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

64 extracted references · 64 canonical work pages · 10 internal anchors

  1. [1]

    and Jeffrey, Kyle and Ruano, Rosario Jauregui and Hsu, Jasmine and Gopalakrishnan, Keerthana and David, Byron and Zeng, Andy and Fu, Chuyuan Kelly , booktitle =

    ichter, brian and Brohan, Anthony and Chebotar, Yevgen and Finn, Chelsea and Hausman, Karol and Herzog, Alexander and Ho, Daniel and Ibarz, Julian and Irpan, Alex and Jang, Eric and Julian, Ryan and Kalashnikov, Dmitry and Levine, Sergey and Lu, Yao and Parada, Carolina and Rao, Kanishka and Sermanet, Pierre and Toshev, Alexander T and Vanhoucke, Vincent ...

    work page 2023

  2. [2]

    LLM -A*: Large Language Model Enhanced Incremental Heuristic Search on Path Planning

    Meng, Silin and Wang, Yiwei and Yang, Cheng-Fu and Peng, Nanyun and Chang, Kai-Wei. LLM -A*: Large Language Model Enhanced Incremental Heuristic Search on Path Planning. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024. doi:10.18653/v1/2024.findings-emnlp.60

    work page doi:10.18653/v1/2024.findings-emnlp.60 2024

  3. [3]

    2024 , address =

    Tabrizian, Amin and Gupta, Pranav and Taye, Abenezer and Jones, James and Thompson, Ellis and Chen, Shulu and Bonin, Timothy and Eberle, Derek and Wei, Peng , booktitle=. 2024 , address =

    work page 2024

  4. [4]

    Wei, Jason and Wang, Xuezhi and Schuurmans, Dale and Bosma, Maarten and ichter, brian and Xia, Fei and Chi, Ed and Le, Quoc V and Zhou, Denny , booktitle =

    work page

  5. [5]

    Can LLMs plan paths with extra hints from solvers?

    Erik Wu and Sayan Mitra , year=. 2410.05045 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv

  6. [6]

    One-Shot Strategically Deconflicted Route and Operational Volume Generation for Urban Air Mobility Operations , doi =

    Thompson, Ellis and Xu, Yan and Wei, Peng , year =. One-Shot Strategically Deconflicted Route and Operational Volume Generation for Urban Air Mobility Operations , doi =

    work page

  7. [7]

    2023 , eprint=

    ReAct: Synergizing Reasoning and Acting in Language Models , author=. 2023 , eprint=

    work page 2023

  8. [8]

    2025 , eprint=

    FM-Planner: Foundation Model Guided Path Planning for Autonomous Drone Navigation , author=. 2025 , eprint=

    work page 2025

  9. [9]

    UAVs meet LLMs: Overviews and perspectives towards agentic low-altitude mobility , volume=

    Tian, Yonglin and Lin, Fei and Li, Yiduo and Zhang, Tengchao and Zhang, Qiyao and Fu, Xuan and Huang, Jun and Dai, Xingyuan and Wang, Yutong and Tian, Chunwei and Li, Bai and Lv, Yisheng and Kovács, Levente and Wang, Fei-Yue , year=. UAVs meet LLMs: Overviews and perspectives towards agentic low-altitude mobility , volume=. doi:10.1016/j.inffus.2025.10315...

    work page doi:10.1016/j.inffus.2025.103158 2025

  10. [10]

    Eighteenth National Conference on Artificial Intelligence , pages =

    Koenig, Sven and Likhachev, Maxim , title =. Eighteenth National Conference on Artificial Intelligence , pages =. 2002 , isbn =

    work page 2002

  11. [11]

    , year =

    Stentz, A. , year =. Optimal and efficient path planning for partially-known environments , isbn =. Proceedings of the 1994. doi:10.1109/ROBOT.1994.351061 , urldate =

    work page doi:10.1109/robot.1994.351061 1994

  12. [12]

    Technological Forecasting and Social Change , author =

    Societal acceptance of urban air mobility based on the technology adoption framework , volume =. Technological Forecasting and Social Change , author =. 2023 , pages =. doi:10.1016/j.techfore.2023.122807 , language =

    work page doi:10.1016/j.techfore.2023.122807 2023

  13. [13]

    FAA aerospace forecast fiscal years 2020-2040

    Federal Aviation Administration. FAA aerospace forecast fiscal years 2020-2040

    work page 2020

  14. [14]

    Edsger Wybe Dijkstra: His Life, Work, and Legacy , pages=

    A note on two problems in connexion with graphs , author=. Edsger Wybe Dijkstra: His Life, Work, and Legacy , pages=

    work page

  15. [15]

    Energy Demand Analysis for

    Abenezer Taye and Peng Wei and James Jones , booktitle=. Energy Demand Analysis for

    work page

  16. [16]

    Abenezer Taye and Shulu Chen and Peng Wei , booktitle=

    work page

  17. [17]

    and Xu, Yan and Wei, Peng , month = jun, year =

    Thompson, Ellis L. and Xu, Yan and Wei, Peng , month = jun, year =. A. 2023. doi:10.1109/ICUAS57906.2023.10156164 , urldate =

    work page doi:10.1109/icuas57906.2023.10156164 2023

  18. [18]

    Tong, Baiming and Liu, Qingbao and Dai, Chaofan , month = dec, year =. A. 2019. doi:10.1109/IAEAC47372.2019.8997746 , abstract =

    work page doi:10.1109/iaeac47372.2019.8997746 2019

  19. [19]

    and LaValle, S.M

    Kuffner, J.J. and LaValle, S.M. , booktitle=. RRT-connect: An efficient approach to single-query path planning , year=

    work page

  20. [20]

    IEEE Transactions on Robotics and Automation , author =

    Probabilistic roadmaps for path planning in high-dimensional configuration spaces , volume =. IEEE Transactions on Robotics and Automation , author =. 1996 , pages =. doi:10.1109/70.508439 , number =

    work page doi:10.1109/70.508439 1996

  21. [21]

    IEEE Transactions on Systems Science and Cybernetics , author =

    A. IEEE Transactions on Systems Science and Cybernetics , author =. 1968 , keywords =. doi:10.1109/TSSC.1968.300136 , abstract =

    work page doi:10.1109/tssc.1968.300136 1968

  22. [22]

    IEEE Access , author =

    Three-. IEEE Access , author =. 2018 , pages =. doi:10.1109/ACCESS.2018.2824558 , urldate =

    work page doi:10.1109/access.2018.2824558 2018

  23. [23]

    Adiyatov, Olzhas and Varol, Huseyin Atakan , month = aug, year =. A novel. 2017. doi:10.1109/ICMA.2017.8016024 , urldate =

    work page doi:10.1109/icma.2017.8016024 2017

  24. [24]

    DiLu: A Knowledge-Driven Approach to Autonomous Driving with Large Language Models

    Wen, Licheng and Fu, Daocheng and Li, Xin and Cai, Xinyu and Ma, Tao and Cai, Pinlong and Dou, Min and Shi, Botian and He, Liang and Qiao, Yu , month = feb, year =. doi:10.48550/arXiv.2309.16292 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2309.16292

  25. [25]

    Zhao, Wayne Xin and Zhou, Kun and Li, Junyi and Tang, Tianyi and Wang, Xiaolei and Hou, Yupeng and Min, Yingqian and Zhang, Beichen and Zhang, Junjie and Dong, Zican and Du, Yifan and Yang, Chen and Chen, Yushuo and Chen, Zhipeng and Jiang, Jinhao and Ren, Ruiyang and Li, Yifan and Tang, Xinyu and Liu, Zikang and Liu, Peiyu and Nie, Jian-Yun and Wen, Ji-R...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2303.18223

  26. [26]

    GPT-4 Technical Report

    2024 , note =. doi:10.48550/arXiv.2303.08774 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2303.08774 2024

  27. [27]

    and Chatterji, Gano Broto and Sheth, Kapil and Lai, Chok Fung and Sridhar, Banavar and Edholm, Kaj-Martin and Erzberger, Heinz , month = jun, year =

    Pradeep, Priyank and Lauderdale, Todd A. and Chatterji, Gano Broto and Sheth, Kapil and Lai, Chok Fung and Sridhar, Banavar and Edholm, Kaj-Martin and Erzberger, Heinz , month = jun, year =. Wind-. doi:10.2514/6.2020-3271 , language =

    work page doi:10.2514/6.2020-3271 2020

  28. [28]

    Journal of Aerospace Information Systems , author =

    Energy-. Journal of Aerospace Information Systems , author =. 2019 , pages =. doi:10.2514/1.I010710 , language =

    work page doi:10.2514/1.i010710 2019

  29. [29]

    , month = sep, year =

    Becerra, Victor M. , month = sep, year =. Solving complex optimal control problems at no cost with. 2010. doi:10.1109/CACSD.2010.5612676 , abstract =

    work page doi:10.1109/cacsd.2010.5612676 2010

  30. [30]

    Driess, Danny and Xia, Fei and Sajjadi, Mehdi S. M. and Lynch, Corey and Chowdhery, Aakanksha and Ichter, Brian and Wahid, Ayzaan and Tompson, Jonathan and Vuong, Quan and Yu, Tianhe and Huang, Wenlong and Chebotar, Yevgen and Sermanet, Pierre and Duckworth, Daniel and Levine, Sergey and Vanhoucke, Vincent and Hausman, Karol and Toussaint, Marc and Greff,...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2303.03378

  31. [31]

    SEED-Bench-2: Benchmarking Multimodal Large Language Models

    Li, Bohao and Ge, Yuying and Ge, Yixiao and Wang, Guangzhi and Wang, Rui and Zhang, Ruimao and Shan, Ying , month = nov, year =. doi:10.48550/arXiv.2311.17092 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2311.17092

  32. [32]

    Voyager: An Open-Ended Embodied Agent with Large Language Models

    Wang, Guanzhi and Xie, Yuqi and Jiang, Yunfan and Mandlekar, Ajay and Xiao, Chaowei and Zhu, Yuke and Fan, Linxi and Anandkumar, Anima , month = oct, year =. Voyager:. doi:10.48550/arXiv.2305.16291 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2305.16291

  33. [33]

    Incremental Sampling-based Algorithms for Optimal Motion Planning

    Karaman, Sertac and Frazzoli, Emilio , month = may, year =. Incremental. doi:10.48550/arXiv.1005.0416 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1005.0416

  34. [34]

    Fu, Daocheng and Li, Xin and Wen, Licheng and Dou, Min and Cai, Pinlong and Shi, Botian and Qiao, Yu , month = jul, year =. Drive. doi:10.48550/arXiv.2307.07162 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2307.07162

  35. [35]

    doi:10.48550/arXiv.2312.09245 , abstract =

    Wang, Wenhai and Xie, Jiangwei and Hu, ChuanYang and Zou, Haoming and Fan, Jianan and Tong, Wenwen and Wen, Yang and Wu, Silei and Deng, Hanming and Li, Zhiqi and Tian, Hao and Lu, Lewei and Zhu, Xizhou and Wang, Xiaogang and Qiao, Yu and Dai, Jifeng , month = dec, year =. doi:10.48550/arXiv.2312.09245 , abstract =

    work page doi:10.48550/arxiv.2312.09245

  36. [36]

    and Solis, Eduardo and Patterson, Michael D

    Silva, Christopher and Johnson, Wayne R. and Solis, Eduardo and Patterson, Michael D. and Antcliff, Kevin R. , month = jun, year =. 2018. doi:10.2514/6.2018-3847 , language =

    work page doi:10.2514/6.2018-3847 2018

  37. [37]

    Journal of Physics: Conference Series , author =

    Conceptual design and sizing optimization based on minimum energy consumption of lift-cruise type. Journal of Physics: Conference Series , author =. 2022 , pages =. doi:10.1088/1742-6596/2385/1/012072 , abstract =

    work page doi:10.1088/1742-6596/2385/1/012072 2022

  38. [38]

    Proceedings of the National Academy of Sciences , author =

    The promise of energy-efficient battery-powered urban aircraft , volume =. Proceedings of the National Academy of Sciences , author =. 2021 , pages =. doi:10.1073/pnas.2111164118 , abstract =

    work page doi:10.1073/pnas.2111164118 2021

  39. [39]

    and Pradeep, Priyank and Chatterji, Gano Broto , month = jun, year =

    Kulkarni, Chetan S. and Pradeep, Priyank and Chatterji, Gano Broto , month = jun, year =. Simulation. doi:10.2514/6.2022-3648 , language =

    work page doi:10.2514/6.2022-3648 2022

  40. [40]

    Interactive

    Reuther, Albert and Kepner, Jeremy and Byun, Chansup and Samsi, Siddharth and Arcand, William and Bestor, David and Bergeron, Bill and Gadepally, Vijay and Houle, Michael and Hubbell, Matthew and Jones, Michael and Klein, Anna and Milechin, Lauren and Mullen, Julia and Prout, Andrew and Rosa, Antonio and Yee, Charles and Michaleas, Peter , month = sep, ye...

    work page doi:10.1109/hpec.2018.8547629 2018

  41. [41]

    and Klemp, Joseph B

    Skamarock, William C. and Klemp, Joseph B. and Dudhia, Jimy and Gill, David O. and Liu, Zhiquan and Berner, Judith and Wang, Wei and Powers, Jordan G. and Duda, Michael G. and Barker, Dale M. and Huang, Xiang-Yu , month = mar, year =. A. doi:10.5065/1DFH-6P97 , note =

    work page doi:10.5065/1dfh-6p97

  42. [42]

    and Bonin, Timothy and Mitchell, Erin , month = jun, year =

    Jones, James C. and Bonin, Timothy and Mitchell, Erin , month = jun, year =. Evaluating. doi:10.2514/6.2023-4104 , language =

    work page doi:10.2514/6.2023-4104 2023

  43. [43]

    Training language models to follow instructions with human feedback , url =

    Ouyang, Long and Wu, Jeffrey and Jiang, Xu and Almeida, Diogo and Wainwright, Carroll and Mishkin, Pamela and Zhang, Chong and Agarwal, Sandhini and Slama, Katarina and Ray, Alex and Schulman, John and Hilton, Jacob and Kelton, Fraser and Miller, Luke and Simens, Maddie and Askell, Amanda and Welinder, Peter and Christiano, Paul F and Leike, Jan and Lowe,...

    work page 2022

  44. [44]

    Chain-of-

    Tabrizian, Amin and Ghazanfari, Mahyar and Wei, Peng , month = jul, year =. Chain-of-. doi:10.2514/6.2025-3711 , language =

    work page doi:10.2514/6.2025-3711 2025

  45. [45]

    Advanced Intelligent Computing Technology and Applications: 21st International Conference, ICIC 2025, Ningbo, China, July 26–29, 2025, Proceedings, Part XXIII , pages =

    Deng, Hourui and Zhang, Hongjie and Ou, Jie and Feng, Chaosheng , title =. Advanced Intelligent Computing Technology and Applications: 21st International Conference, ICIC 2025, Ningbo, China, July 26–29, 2025, Proceedings, Part XXIII , pages =. 2025 , isbn =. doi:10.1007/978-981-95-0014-7_1 , abstract =

    work page doi:10.1007/978-981-95-0014-7_1 2025

  46. [46]

    2026 , url=

    Zhiwen Chen and Hanming Deng and Zhuoren Li and Huanxi Wen and Guizhe Jin and Ran Yu and Bo Leng , booktitle=. 2026 , url=

    work page 2026

  47. [47]

    Liu, Bo and Jiang, Yuqian and Zhang, Xiaohan and Liu, Qiang and Zhang, Shiqi and Biswas, Joydeep and Stone, Peter , journal=

    work page

  48. [48]

    2024 , url=

    Gautier Dagan and Frank Keller and Alex Lascarides , booktitle=. 2024 , url=

    work page 2024

  49. [49]

    What's the Plan? Evaluating and Developing Planning-Aware Techniques for Language Models

    Eran Hirsch and Guy Uziel and Ateret Anaby-Tavor , year=. 2402.11489 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv

  50. [50]

    2024 IEEE International Conference on Robotics and Automation (ICRA) , pages=

    Isr-llm: Iterative self-refined large language model for long-horizon sequential task planning , author=. 2024 IEEE International Conference on Robotics and Automation (ICRA) , pages=. 2024 , organization=

    work page 2024

  51. [51]

    Gonzalez and Ion Stoica , booktitle=

    Lianmin Zheng and Wei-Lin Chiang and Ying Sheng and Siyuan Zhuang and Zhanghao Wu and Yonghao Zhuang and Zi Lin and Zhuohan Li and Dacheng Li and Eric Xing and Hao Zhang and Joseph E. Gonzalez and Ion Stoica , booktitle=. 2023 , url=

    work page 2023

  52. [52]

    LaValle, Steven M , year=

    work page

  53. [53]

    and LaValle, S.M

    Kuffner, J.J. and LaValle, S.M. , booktitle=. 2000 , volume=

    work page 2000

  54. [54]

    Mission Planner Overview , author =

    work page

  55. [55]

    2025 , volume=

    Wang, Wenhao and Li, Yanyan and Jiao, Long and Yuan, Jiawei , booktitle=. 2025 , volume=

    work page 2025

  56. [56]

    2025 , volume=

    Chen, Guojun and Yu, Xiaojing and Ling, Neiwen and Zhong, Lin , journal=. 2025 , volume=. doi:10.1109/TMC.2025.3561282 , url =

    work page doi:10.1109/tmc.2025.3561282 2025

  57. [57]

    Proceedings of the 32nd ACM International Conference on Multimedia , pages =

    Xu, Weiye and Wang, Min and Zhou, Wengang and Li, Houqiang , title =. Proceedings of the 32nd ACM International Conference on Multimedia , pages =. 2024 , isbn =. doi:10.1145/3664647.3680661 , abstract =

    work page doi:10.1145/3664647.3680661 2024

  58. [58]

    The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

    Kaiqu Liang and Zixu Zhang and Jaime Fern. The Thirty-eighth Annual Conference on Neural Information Processing Systems , year=

    work page

  59. [59]

    2025 , url=

    Marius Memmel and Jacob Berg and Bingqing Chen and Abhishek Gupta and Jonathan Francis , booktitle=. 2025 , url=

    work page 2025

  60. [60]

    2025 , eprint =

    Agentic. 2025 , eprint =

    work page 2025

  61. [61]

    2025 , url =

    Anonymous , booktitle =. 2025 , url =

    work page 2025

  62. [62]

    and Smith, Leslie N

    Comisky, Tyler D. and Smith, Leslie N. and Roberts, Mark and Lovejoy, Joshua and Garg, Avni and Nam, Luke and others , booktitle =. Towards Siloed. 2025 , url =

    work page 2025

  63. [63]

    A Roadmap to Guide the Integration of

    Anonymous , booktitle =. A Roadmap to Guide the Integration of. 2025 , url =

    work page 2025

  64. [64]

    Guo, Daya and Yang, Dejian and Zhang, Haowei and Song, Junxiao and Wang, Peiyi and Zhu, Qihao and Xu, Runxin and Zhang, Ruoyu and Ma, Shirong and Bi, Xiao and Zhang, Xiaokang and Yu, Xingkai and Wu, Yu and Wu, Z. F. and Gou, Zhibin and Shao, Zhihong and Li, Zhuoshu and Gao, Ziyi and Liu, Aixin and Xue, Bing and Wang, Bingxuan and Wu, Bochao and Feng, Bei ...

    work page doi:10.1038/s41586-025-09422-z