Pith.

Artificial Intelligence · arXiv:2611.24004 v1 · submitted 18 Nov 2026 · 49 references

Space-X: Self-Evolving Spatial Agents Weaving Clues into Evidence Chains

Zijian Zhang, Wenqing Tian, Qiang Liu, Liang Wang

The paper says

“Spatial reasoning connects visual observations, geometric estimates, and linguistic decisions into a coherent derivation. Yet tool-augmented agents often acquire evidence without clear targets, combine outputs under inconsistent object or reference-frame semantics, and misjudge evidence sufficiency.”

more of the abstract

Thus, more tools or calls do not reliably improve reasoning. We introduce Space-X, a spatial-reasoning framework that self-evolves toward complete, continuous evidence-chain construction. An Execution Agent acquires, associates, and judges evidence per problem using role-specific memory. A Control Agent diagnoses evidence-chain breaks from execution traces and jointly revises tools, interfaces, skills, memory organization, and decision rules. This loop co-evolves capability and evidence use: evidence needs guide capability development; usable results guide actions. Across multiple benchmarks, Space-X yields gains with four base models. Relative to baselines, GPT-6-sol improves answer accuracy on ViewSpatial and MMSI by 50.39% and 23.07%, respectively; Gemini-3.8-Flash gains 47.50% and 25.22% on these benchmarks. This provides a unified route from fragmented spatial clues to traceable, derivable answers.

from the abstract

Pith says

Machine review · deepseek-v4-flash · 19 Nov 2026 · read the full text (166,178 characters) · not checked by a person

It can be wrong. Challenge this review or reply to it.

Holds up, with conditions. Space-X builds a self-evolving spatial reasoning framework that jointly optimizes tools and evidence-use rules, reporting large gains on multiple benchmarks; the core idea is novel and the empirical results are impressive, though the reliance on a proprietary LLM for the control agent and incomplete comparisons to the strongest baselines temper confidence. This line is wrong

10 minor · 1 open question

How often objections from reviews like this one are right: not yet measured

Are you an author? What you can do.

The paper, first character to last

The reviewer was given the whole text, 166,178 characters. Not checked: figures, code, what the 49 references say.

10 minor objections

The paper says, at Abstract

Pith says: minor objection 1 of 10

The abstract states 'Relative to baselines, GPT-6-sol improves answer accuracy on ViewSpatial and MMSI by 50.39% and 23.07%, respectively.' The phrase 'relative to baselines' is ambiguous: it could mean relative to the base model (GPT-6-sol alone) or relative to all baselines. From Table 1, the 50.39% is relative to GPT-6-sol alone (53.19 to 79.99). Clarify this in the abstract.

The paper says, at §4.2, Table 2

Pith says: minor objection 2 of 10

In Table 2, the row for S-Agent 2.0 (Qwen3.8-27B) shows a score of 66.50 on MMSI Overall. Space-X (GPT-6-sol) scores 66.33, which is slightly lower. The paper should comment on this directly rather than only stating that Space-X 'exceeds S-Agent 2.0 on MMSI' (which is true for GPT-6-astra and Gemini-4-Argon but not for GPT-6-sol).

show 8 more

The paper says, at §3.2, Eq. 4

Pith says: minor objection 3 of 10

Equation 4 defines task constraints C_x = (R, F, ρ, Ω) and evidence items e_i = (v_i, σ_i, ρ_i, η_i). The symbol ρ is used for both the queried relation in C_x and the semantic binding in e_i. While the context distinguishes them, using different symbols would improve clarity.

The paper says, at §4.1, Table 1

Pith says: minor objection 4 of 10

Table 1 reports VSI results with 'Only 10 frames' but does not specify how the 10 frames are selected (random, uniform, or by the method). The VSI results in Appendix Table A.8 provide more detail, but the main text should briefly note the selection strategy.

The paper says, at §4.3, Table 4

Pith says: minor objection 5 of 10

Table 4 reports 'Mismatch ↓' and 'Delivery ↑' as process metrics but does not define them in the main text. They are defined in Appendix §A.6.4, but a brief definition in the main text would help readers interpret the table without flipping to the appendix.

The paper says, at §4.4, Table 6

Pith says: minor objection 6 of 10

Table 6 shows X-SFT results after 15 iterations. The paper should specify whether the 15 iterations are the same as the R=15 in Algorithm 1 (Appendix §A.2.4) or additional iterations on top of the original construction.

The paper says, at §4.4, Table 7

Pith says: minor objection 7 of 10

Table 7 compares general harnesses (DSH CLI, Codex CLI) with and without spatial tools. The 'Calls/question' column averages spatial calls per question equally across benchmarks. It would be helpful to report the breakdown per benchmark, as the number of spatial calls may vary significantly between ViewSpatial, SPAR, and MMSI.

The paper says, at Appendix §A.8, Eq. A.10

Pith says: minor objection 8 of 10

The acceptance threshold equation (Eq. A.10) requires improvement strictly greater than zero in at least one target category and zero regression across all categories. This is a very strict criterion. The paper should discuss whether this threshold was actually used during development or is a recommended protocol for future work.

The paper says, at §2 (Related Work)

Pith says: minor objection 9 of 10

The related work section is comprehensive but would benefit from a brief discussion of how Space-X differs from the cited self-improvement approaches (ADAS, Darwin Gödel Machine, EvoSpatial, HiMeco, SHAPER) in terms of the specific mechanism of evidence-chain-driven evolution rather than just stating that 'task scores do not localize failures.'

The paper says, at §3.3, Eq. 7

Pith says: minor objection 10 of 10

Equation 7 defines the missing sources D_t^{img} = R_t^{img} \ A_t^{img}. The notation uses set difference, but the paper does not specify how images are identified (by source ID, filename, or content hash). Clarify the identity criterion for images.

Open questions (1)

Pith raised these and could not confirm them against the text of the paper it read. They are questions for the authors, not findings.

  1. §3.1, Eq. 3 and surrounding text Pith raised a concern here that it could not check against the text of the paper it was given, so it is not published as an objection.

What would settle it

If a human expert reviewed a sample of the Control Agent's diagnoses and found that a majority misidentified the root cause of the evidence-chain break (e.g., attributing a failure to missing tools when it was actually due to incorrect memory retrieval), the central claim that self-evolution driven by these diagnoses improves reasoning would be undermined.

from Pith's summary of the paper

How the review went

5 passes by deepseek-v4-flash · 19 Nov 2026

  1. Reader holds up, with conditions confidence moderate
  2. Skeptic concern withdrawn: Pith's reviewer misread the paper 9 s · 46,936 tokens read, 792 written
  3. Referee 10 minor 21 s · 47,320 tokens read, 2,897 written
  4. Check against the paper kept 0 of 6 objections 32 s · 96,503 tokens read, 5,603 written
  5. Editor wrote the verdict above 7 s · 47,700 tokens read, 569 written

Not run on this paper: the simulated authors' defence.

Pith's passes raised 6 objections. A check against the paper's text kept 0, withdrew 5 (5 Pith's reviewer misread the paper) and left 1 as a question.

The review in full

The review in brief, machine-written

This paper introduces Space-X, a dual-agent framework for spatial reasoning that couples an Execution Agent (which solves tasks using role-specific memory and tool calls) with a Control Agent (which diagnoses evidence-chain breaks from execution traces and jointly revises tools, interfaces, memory organization, and decision rules). The central idea—making evidence-chain continuity the objective of self-evolution—is a solid new contribution. The reported gains are large: GPT-6-sol improves accuracy on ViewSpatial by 50.39% and on MMSI by 23.07% relative to baselines, and the framework improves four different base models across multiple benchmarks. The ablation studies are informative, showing that joint evolution of tools and evidence rules outperforms evolving either alone, and that explicit evidence association with memory continuity substantially reduces semantic mismatches and improves evidence delivery. The qualitative cases and detailed appendix provide a thorough account of how the framework works in practice.

That said, several factors lower confidence. The comparison to the strongest prior spatial agent, S-Agent 2.0, is limited to MMSI only; results on ViewSpatial and SPAR are not reported for that baseline, making it hard to assess the relative improvement. The Control Agent relies on GPT-6-sol (a proprietary model) for diagnosis and editing, raising reproducibility concerns—the paper does not provide a formal guarantee or human verification that the self-evolution loop converges to a correct or optimal harness. The paper acknowledges that task scores do not localize failures but assumes the richer trace feedback suffices; this assumption is not independently validated. The free parameters (e.g., number of API calls, token limits, context character limits) are numerous and their sensitivity is not explored. The paper also does not release code or data, which would be essential for independent reproduction.

Overall, the paper presents a novel framework with strong empirical results that deserve serious peer review. The authors should be asked to provide missing baseline comparisons (especially S-Agent 2.0 on ViewSpatial and SPAR), release code and data, and discuss the limitations of relying on a proprietary LLM for the control loop. The central claim is plausible but requires stronger evidence.

Recommendation: Send to peer review. The paper is substantial enough to warrant expert scrutiny, and the authors should be given the opportunity to address the concerns.

The stress test's concern

Withdrawn: Pith's reviewer misread the paper.

The paper says

“Development uses SAT-v2~\cite{ray2024sat}, SIMS-VSI~\cite{brown2025simsv}, NVIDIA Warehouse~\cite{tang2025aicity}, and GRiD-3D~\cite{lee2022grid3d}, covering perspective changes, relations across images, quantitative geometry, and local visual discrimination.”

The referee's report

Not published. Pith's referee wrote this assessment before its objections were checked against the paper, and 5 of its 5 major objections did not survive that check.

Pith's reading of the paper (a machine's summary, not the authors')

The paper introduces Space-X, a framework for spatial reasoning that uses two agents: an Execution Agent that solves tasks by building evidence chains, and a Control Agent that diagnoses breaks in those chains and revises the tools, memory, and rules accordingly. The central claim is that this self-evolution loop, driven by evidence-chain continuity, substantially improves spatial reasoning accuracy across multiple benchmarks and base models. The authors report that Space-X with GPT-6-sol improves answer accuracy on ViewSpatial by 50.39% and on MMSI by 23.07% relative to baselines. The paper argues that the key to these gains is not simply adding more tools, but co-evolving capabilities with the rules for selecting, interpreting, associating, and judging evidence.

What carries the argument. The central mechanism is the dual-agent self-evolution loop. The Execution Agent (A_e) constructs an evidence chain for each task using role-specific memory (Working, Observation, Spatial, Episodic, Procedural). The Control Agent (A_c) diagnoses evidence-chain breaks from execution traces and proposes edits to the harness (tools, interfaces, skills, memory organization, decision rules). The loop co-evolves capability and evidence use: evidence needs guide capability development, and usable results guide actions. The reliability of an evidence chain is defined by three conditions: coverage of task premises, semantic alignment of evidence, and traceable provenance.

Core claim. The paper claims that spatial reasoning in tool-augmented agents is limited not by the number of tools available, but by breaks in the evidence chain that connects observations to answers. These breaks arise from untargeted evidence acquisition, inconsistent evidence interpretation (mixing results from different objects, frames, or sources), and misjudged evidence sufficiency. The authors propose that a dual-agent self-evolution loop, where a Control Agent diagnoses evidence-chain breaks from execution traces and jointly revises tools, interfaces, memory organization, and decision rules, can systematically close these gaps. The resulting framework, Space-X, is claimed to produce substantial gains across multiple benchmarks and base models by ensuring that evidence needs guide capability development and usable results guide the next action.

The premise it rests on. The Control Agent's diagnoses and edits, generated by a large language model from execution traces, reliably identify the true cause of evidence-chain breaks and produce structural changes that generalize to new tasks.

If the paper is right (Pith's inferences, not the authors')

If the framework's claims hold, spatial reasoning agents could be built that improve autonomously from execution feedback, reducing the need for manual engineering of tool-use policies.

The approach suggests that the bottleneck in tool-augmented reasoning is not tool availability but the coherence of the evidence chain, implying that future work should focus on diagnosis and repair of reasoning structure rather than expanding tool libraries.

The reported transfer of the harness across different base models (GPT-6-sol, GPT-6-astra, Gemini-3.8-Flash, Gemini-4-Argon) suggests that the evolved evidence-chain structure generalizes beyond the model used during development.

The ablation studies indicate that jointly evolving tools and evidence-use rules is more effective than evolving either in isolation, which could guide the design of other self-improving agent systems.

The paper's reliance on GPT-6-sol for both agent roles during initial construction means the framework's effectiveness may be partly dependent on the diagnostic and generative capabilities of that specific model; it is an open question how well the approach would work with weaker base models as the control agent.

The paper does not provide a formal guarantee that the self-evolution loop converges to an optimal or correct harness; the diagnosis step relies on the LLM's ability to localize failures from traces, which could be brittle or miss root causes. A testable extension would be to compare the LLM-generated diagnoses against human-annotated failure analyses on a held-out set.

The framework's emphasis on evidence-chain continuity could be applied to other reasoning domains beyond spatial reasoning, such as temporal or causal reasoning, where fragmented evidence and misaligned semantics are also common failure modes.

Check this review

What was read

The passes were given all 166,178 characters of the paper's LaTeX source.

Fingerprint of that text (SHA-256): c57339d87afe454db0690fb5637d920f4d55c6b70df8cf28e585a8da9ce26943

Pith does not keep that text. Rebuilt from the paper's source or PDF by the same code, it has this fingerprint if it is the same text. From a copy of Pith's code:

PYTHONPATH=. python scripts/review_input_check.py 2611.24004

The passes, in the order they ran

  1. Reader deepseek-v4-flash · medium effort · $0.0050 · instructions of the Reader, fingerprint f9ead7300382…
  2. Skeptic deepseek-v4-flash · medium effort · $0.0044 · instructions of the Skeptic, fingerprint 8f88c0f61488…
  3. Referee deepseek-v4-flash · medium effort · $0.0048 · instructions of the Referee, fingerprint a852d99aea0d…
  4. Check against the paper deepseek-v4-flash · medium effort · $0.0063 · instructions of the Check against the paper, fingerprint 6c7a6883d551…
  5. Editor deepseek-v4-flash · medium effort · $0.0044 · instructions of the Editor, fingerprint e7211f6a3d27…
  6. Circularity check deepseek-v4-flash · medium effort · $0.0043 · instructions of the Circularity check, fingerprint e9946e39d67b…
  7. Summary for the page deepseek-v4-flash · medium effort · temperature 0.2 · $0.0047 · instructions of the Summary for the page, fingerprint e4f56bf568e7…

The passes cost $0.034 in model calls together.

Version stamped on this review: pith-pipeline@v1.4.1-daily-deepseek-or-medium+deepseek/deepseek-v4-flash+p.d60f54226f00

Each link opens a pass's instructions as plain text, the exact bytes its fingerprint was taken over. Every stored set of instructions.

curl -s https://pith.science/methodology/prompts/f9ead730038216984cfb56d7cb34332bdf691828296a1de4245f3d6667d4bf56 | sha256sum

prints the fingerprint of the reader's instructions. The same works for every pass.

The receipt

The review's fingerprint was taken on 19 Nov 2026. The review you are reading still matches it.

Fingerprint of the review (SHA-256): 91bbee0dac5a8392a28eb43f52cc3102f62195a8dc28a9df9213c1f62fb6caf8

curl -s 'https://pith.science/api/v1/verdicts/2611.24004?form=canonical' | sha256sum

prints the fingerprint of the review as it stands now. The rule, in four steps.

A matching fingerprint shows the review has not changed since that date. It does not show that the review is right.

How often reviews like this one are right

Reviews like this one: written by deepseek-v4-flash, from the paper's LaTeX source.

How often objections from reviews like this one are right: not yet measured

Every value above, in the review's public record (JSON)

Reviews by people

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

Spatial reasoning connects visual observations, geometric estimates, and linguistic decisions into a coherent derivation. Yet tool-augmented agents often acquire evidence without clear targets, combine outputs under inconsistent object or reference-frame semantics, and misjudge evidence sufficiency. Thus, more tools or calls do not reliably improve reasoning. We introduce Space-X, a spatial-reasoning framework that self-evolves toward complete, continuous evidence-chain construction. An Execution Agent acquires, associates, and judges evidence per problem using role-specific memory. A Control Agent diagnoses evidence-chain breaks from execution traces and jointly revises tools, interfaces, skills, memory organization, and decision rules. This loop co-evolves capability and evidence use: evidence needs guide capability development; usable results guide actions. Across multiple benchmarks, Space-X yields gains with four base models. Relative to baselines, GPT-6-sol improves answer accuracy on ViewSpatial and MMSI by 50.39% and 23.07%, respectively; Gemini-3.8-Flash gains 47.50% and 25.22% on these benchmarks. This provides a unified route from fragmented spatial clues to traceable, derivable answers.

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-11-19

The authors' main claim, as the classifier picked it out 0.48

“This provides a unified route from fragmented spatial clues to traceable, derivable answers.”

How the main claim is stated: asserted 0.98

What the abstract itself states, as probabilities: claims first 0.07 · states limitation 0.13 · human participants 0.03 · controlled comparison 0.92 · randomized assignment 0.02 · states quantitative result 0.99 · compares with prior results 0.98 · mentions code or data availability 0.03

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
Spatial reasoning connects visual observations, geometric estimates, and linguistic decisions into a coherent derivation. background 1.00 0.07 0.04 asserted 1.00
Yet tool-augmented agents often acquire evidence without clear targets, combine outputs under inconsistent object or reference-frame semantics, and misjudge evidence sufficiency. gap 0.54 0.09 0.04 asserted 0.99
Thus, more tools or calls do not reliably improve reasoning. background 0.63 0.24 0.11 asserted 0.94
We introduce Space-X, a spatial-reasoning framework that self-evolves toward complete, continuous evidence-chain construction. contribution 1.00 0.87 0.19 asserted 1.00
An Execution Agent acquires, associates, and judges evidence per problem using role-specific memory. method 0.94 0.72 0.05 asserted 1.00
A Control Agent diagnoses evidence-chain breaks from execution traces and jointly revises tools, interfaces, skills, memory organization, and decision rules. method 0.95 0.70 0.07 asserted 0.99
This loop co-evolves capability and evidence use: evidence needs guide capability development; usable results guide actions. method 0.85 0.57 0.16 asserted 1.00
Across multiple benchmarks, Space-X yields gains with four base models. result 0.98 0.95 0.23 asserted 1.00
Relative to baselines, GPT-6-sol improves answer accuracy on ViewSpatial and MMSI by 50.39% and 23.07%, respectively; Gemini-3.8-Flash gains 47.50% and 25.22% on these benchmarks. result 1.00 0.94 0.15 asserted 1.00
This provides a unified route from fragmented spatial clues to traceable, derivable answers. interpretation 0.97 0.61 0.44 asserted 0.98

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.81

What was examined, and what was not

Sources read

  • arXiv listing abstract v1

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-19: 1 request, 7,429 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

Figures (4)
Figure 1 of the paper
Figure 1. Three recurring evidence-chain breaks and the corresponding S
Figure 2 of the paper
Figure 2. Dual-agent self-evolution in SPACE-X. The Execution Agent constructs an evidence chain for each task; execution feedback lets the Control Agent diagnose broken links and jointly revise tools and evidence-use rules for the next iteration. cover the question’s required premises, consistently bind objects, sources, and reference frames, and make final deci￾sions traceable to explicit support. The Execution Agent Ae solves the task; the Control Agent Ac improves the execu￾tion structure from execution traces and task-level evalua￾tion feedback. Let Cx denote current task constraints and…
Figure 3 of the paper
Figure 3. Task-time execution in SPACE-X. The API selects source-aware tools [11–21], constructs semantically bound evidence, ex￾changes it through five memory roles, and resolves disputed premises with source-linked review before answering. endpoint mismatch; valid results absent from final claims indicate memory-relay or context-delivery failure. Diagno￾sis and editing proceed as Dr = Diagnose(Ac, H(r) , Fr), ∆r = Propose(Ac, H(r) , Dr), He(r) = Edit(Ac, H(r) , ∆r). (5) Each edit addresses an evidence-link failure and speci￾fies capability, source semantics, interpretation conditions, memo
Figure 4 of the paper
Figure 4. Optimization activity and benchmark performance dur

Figures are the authors', reproduced from the paper.

References (49)
  1. [1] Shi-Yu Tian, Zhuo-Xia Wang, Xuan-Yi Zhu, Zhi Zhou, Xinwei Yang, Kun-Yang Yu, Ming Yang, Yang Chen, and Yu-Feng Li. Self-Evolving Neuro-Symbolic Skills for Tool-Augmented Spatial Reasoning.arXiv preprint arXiv:2608.07955, 2026 on PithPith's reviewarXiv
  2. [2] Bin Yu, Shijie Lian, Zhaolong Shen, Xiaopeng Lin, Yao Zhang, Haishan Liu, Haotian Li, Yuliang Wei, Kai Hu, Yichao Du, Zhirui Zhang, Bailing Wang, and Kai Chen. EvoSpatial: Experience-Driven Skill and Tool Evolution for Spatial Reasoning Agents.arXiv preprint arXiv:2610.16679, 2026 on PithPith's reviewarXiv
  3. [3] Haokai Zhang, Yuhang Ding, Yunshu Zhou, Xinze Du, Shengtao Zhang, Zhiyue Zhao, Yuling Xi, and Hao Chen. Spatial Memory Agent: Experience-Grounded Procedu- ral Memory for Spatial Intelligence.arXiv preprint arXiv:2608.12743, 2026 on PithPith's reviewarXiv
  4. [4] Rui Gong, Ruoyu Wang, Yalun Dai, Hao Li, Sidi Yang, Qihao Zhao, Hongyu Li, Zeyang Guan, Fangzhou Hong, Zhaoxi Chen, Kim-Hui Yap, and Ziwei Liu. S-Agent 2.0: Spatial Reasoning with a General Coding Agent.arXiv preprint arXiv:2610.28354, 2026 on PithPith's reviewarXiv
  5. [5] Yike Wu, Keyan Chen, Qixiong Wang, Xi Xiang, Maofei Que, Xiaolong Jiang, Xiaoqin Zhang, Ling Shao, and Shijian Lu. ReTAC: Residual Test-Time Agentic Computation for Video Spatial Reasoning.arXiv preprint arXiv:2610.16315, 2026 on PithPith's reviewarXiv
  6. [6] Yifan Zhou, Ziyang Gong, Xiaojun Ye, Yifei Liu, Donggang Li, Xiaolong Zhou, Zhihang Zhong, and Xue Yang. Spatial- Brain: Spatial Intelligence Growing from Experience.arXiv preprint arXiv:2610.29317, 2026 on PithPith's reviewarXiv
  7. [7] Shengran Hu, Cong Lu, and Jeff Clune. Automated Design of Agentic Systems.arXiv preprint arXiv:2408.08435, 2024 on PithPith's reviewarXiv
  8. [8] Jenny Zhang, Shengran Hu, Cong Lu, Robert Lange, and Jeff Clune. Darwin Gödel Machine: Open-Ended Evolution of Self-Improving Agents.arXiv preprint arXiv:2505.22954, 2025 on PithPith's reviewarXiv
  9. [9] Xin Li, Xiaohuan Gu, Yifan Wang, Shaobin Ling, Tianli Shi, Jingao Xu, Zhimeng Yin, Dapeng Wu, and Yun Wang. Reflect After Failure: Brain-Inspired Self-Evolving Agents for Embodied Spatial Intelligence.arXiv preprint arXiv:2610.20312, 2026 on PithPith's reviewarXiv
  10. [10] Peidong Wang, Zhiming Ma, Ying Chang, Xufang Luo, Yiqun Zhang, Zihan Wang, Xiaocui Yang, Shi Feng, Yuqing Yang, and Dongsheng Li. Self-Evolving Embod- ied Agents via Skill-Harness Evolution.arXiv preprint arXiv:2608.11350, 2026 on PithPith's reviewarXiv
  11. [11] Nicolas Carion, Laura Gustafson, Yuan-Ting Hu, Shoub- hik Debnath, Ronghang Hu, Didac Suris, Chaitanya Ryali, Kalyan Vasudev Alwala, Haitham Khedr, Andrew Huang, Jie Lei, Tengyu Ma, Baishan Guo, Arpit Kalla, Markus Marks, Joseph Greer, Meng Wang, Peize Sun, Roman Rä- dle, Tria... on PithPith's reviewarXiv
  12. [12] Haotong Lin, Sili Chen, Junhao Liew, Donny Y . Chen, Zhenyu Li, Guang Shi, Jiashi Feng, and Bingyi Kang. Depth Anything 3: Recovering the Visual Space from Any Views. arXiv preprint arXiv:2511.10647, 2025 on PithPith's reviewarXiv
  13. [13] Jianyuan Wang, Minghao Chen, Nikita Karaev, Andrea Vedaldi, Christian Rupprecht, and David Novotny. VGGT: Visual Geometry Grounded Transformer.arXiv preprint arXiv:2503.11651, 2025 on PithPith's reviewarXiv
  14. [14] Nikhil Keetha, Norman Müller, Johannes Schönberger, Lorenzo Porzi, Yuchen Zhang, Tobias Fischer, Arno Knapitsch, Duncan Zauss, Ethan Weber, Nelson Antunes, Jonathon Luiten, Manuel Lopez-Antequera, Samuel Rota Bulò, Christian Richardt, Deva Ramanan, Sebastian Scherer, and Peter... on PithPith's reviewarXiv
  15. [15] Ruicheng Wang, Sicheng Xu, Yue Dong, Yu Deng, Jianfeng Xiang, Zelong Lv, Guangzhong Sun, Xin Tong, and Jiaolong Yang. MoGe-2: Accurate Monocular Geometry with Metric Scale and Sharp Details.arXiv preprint arXiv:2507.02546, 2025 on PithPith's reviewarXiv
  16. [16] Shilong Liu, Zhaoyang Zeng, Tianhe Ren, Feng Li, Hao Zhang, Jie Yang, Qing Jiang, Chunyuan Li, Jianwei Yang, Hang Su, Jun Zhu, and Lei Zhang. Grounding DINO: Marry- ing DINO with Grounded Pre-Training for Open-Set Object Detection.arXiv preprint arXiv:2303.05499, 2023 on PithPith's reviewarXiv
  17. [17] Alexander Kirillov, Eric Mintun, Nikhila Ravi, Hanzi Mao, Chloe Rolland, Laura Gustafson, Tete Xiao, Spencer Whitehead, Alexander C. Berg, Wan-Yen Lo, Piotr Dol- lár, and Ross Girshick. Segment Anything.arXiv preprint arXiv:2304.02643, 2023 on PithPith's reviewarXiv
  18. [18] Lihe Yang, Bingyi Kang, Zilong Huang, Zhen Zhao, Xiao- gang Xu, Jiashi Feng, and Hengshuang Zhao. Depth Any- thing V2.arXiv preprint arXiv:2406.09414, 2024 on PithPith's reviewarXiv
  19. [19] Michael Tschannen, Alexey Gritsenko, Xiao Wang, Muham- mad Ferjad Naeem, Ibrahim Alabdulmohsin, Nikhil Parthasarathy, Talfan Evans, Lucas Beyer, Ye Xia, Basil Mustafa, Olivier Hénaff, Jeremiah Harmsen, Andreas Steiner, and Xiaohua Zhai. SigLIP 2: Multilingual Vision- Language ... on PithPith's reviewarXiv
  20. [20] Zehan Wang, Ziang Zhang, Tianyu Pang, Chao Du, Heng- shuang Zhao, and Zhou Zhao. Orient Anything: Learning Robust Object Orientation Estimation from Rendering 3D Models.arXiv preprint arXiv:2412.18605, 2024 on PithPith's reviewarXiv
show 29 more
  1. [21] Xitong Yang, Devansh Kukreja, Don Pinkus, Anushka Sagar, Taosha Fan, Jinhyung Park, Soyong Shin, Jinkun Cao, Jiawei 9 Liu, Nicolas Ugrinovic, Matt Feiszli, Jitendra Malik, Piotr Dollar, and Kris Kitani. SAM 3D Body: Robust Full-Body Human Mesh Recovery.arXiv preprint arXiv:260... on PitharXiv
  2. [22] Zhongang Cai, Yubo Wang, Qingping Sun, Ruisi Wang, Chenyang Gu, Wanqi Yin, Zhiqian Lin, Zhitao Yang, Chen Wei, Oscar Qian, Hui En Pang, Xuanke Shi, Kewang Deng, Xiaoyang Han, Zukai Chen, Jiaqi Li, Xiangyu Fan, Hanming Deng, Lewei Lu, Bo Li, Ziwei Liu, Quan Wang, Dahua Lin, and... on PitharXiv
  3. [23] Dingming Li, Hongxing Li, Zixuan Wang, Yuchen Yan, Hang Zhang, Siqi Chen, Guiyang Hou, Shengpei Jiang, Wenqi Zhang, Yongliang Shen, Weiming Lu, and Yueting Zhuang. ViewSpatial-Bench: Evaluating Multi-perspective Spatial Localization in Vision-Language Models.arXiv preprint arX... on PitharXiv
  4. [24] Hongxing Li, Dingming Li, Zixuan Wang, Yuchen Yan, Hang Wu, Wenqi Zhang, Yongliang Shen, Weiming Lu, Jun Xiao, and Yueting Zhuang. SpatialLadder: Progressive Training for Spatial Reasoning in Vision-Language Models. arXiv preprint arXiv:2510.08531, 2025 on PitharXiv
  5. [25] Yalun Dai, Hao Li, Shulin Tian, Runmao Yao, Yuhao Dong, Fangzhou Hong, Zhaoxi Chen, Fangfu Liu, Tao Wang, Kim-Hui Yap, and Ziwei Liu. S-Agent: Spatial Tool-Use Elicits Reasoning for Spatial Intelligence.arXiv preprint arXiv:2606.20515, 2026 on PithPith's reviewarXiv
  6. [26] Yifei Liu, Xiaolong Zhou, Ziyang Gong, Yifan Zhou, Yuanyuan Gao, Yuning Gong, Yuchen Yang, Hongjie Zhang, Xue Yang, Xiao Sun, and Zhihang Zhong. DeSpatial: The devil of Agentic Spatial Reasoning Lies in 3D Itself.arXiv preprint arXiv:2610.20990, 2026 on PithPith's reviewarXiv
  7. [27] Seokju Cho, Ryo Hachiuma, Abhishek Badki, Hang Su, Byung-Kwan Lee, Chan Hee Song, Sifei Liu, Subhashree Radhakrishnan, Seungryong Kim, Yu-Chiang Frank Wang, and Min-Hung Chen. SpatialClaw: Rethinking Action In- terface for Agentic Spatial Reasoning. Technical report, NVIDIA, 2026 on Pith
  8. [28] Arijit Ray, Jiafei Duan, Ellis Brown, Reuben Tan, Dina Bashkirova, Rose Hendrix, Kiana Ehsani, Aniruddha Kem- bhavi, Bryan A. Plummer, Ranjay Krishna, Kuo-Hao Zeng, and Kate Saenko. SAT: Dynamic Spatial Aptitude Train- ing for Multimodal Language Models.arXiv preprint arXiv:24... on PitharXiv
  9. [29] Ellis Brown, Arijit Ray, Ranjay Krishna, Ross Girshick, Rob Fergus, and Saining Xie. SIMS-V: Simulated Instruction- Tuning for Spatial Video Understanding.arXiv preprint arXiv:2511.04668, 2025 on PitharXiv
  10. [30] Zheng Tang, Shuo Wang, David C. Anastasiu, Ming- Ching Chang, Anuj Sharma, Quan Kong, Norimasa Ko- bori, Munkhjargal Gochoo, Ganzorig Batnasan, Munkh- Erdene Otgonbold, Fady Alnajjar, Jun-Wei Hsieh, Tomasz Kornuta, Xiaolong Li, Yilin Zhao, Han Zhang, Subhashree Radhakrishnan, ... on PithPith's reviewarXiv
  11. [31] Jae Hee Lee, Matthias Kerzel, Kyra Ahrens, Cornelius We- ber, and Stefan Wermter. What is Right for Me is Not Yet Right for You: A Dataset for Grounding Relative Directions via Multi-Task Learning.arXiv preprint arXiv:2205.02671, 2022 on PithPith's reviewarXiv
  12. [32] Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, Wenbin Ge, Zhifang Guo, Qidong Huang, Jie Huang, Fei Huang, Binyuan Hui, Shutong Jiang, Zhao- hai Li, Mingsheng Li, Mei Li, Kaixin Li, Zicheng Lin, Jun... on PithPith's reviewarXiv
  13. [33] Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhao- hai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, H... on PithPith's reviewarXiv
  14. [34] Zhongang Cai, Ruisi Wang, Chenyang Gu, Fanyi Pu, Junx- iang Xu, Yubo Wang, Wanqi Yin, Zhitao Yang, Chen Wei, Qingping Sun, Tongxi Zhou, Jiaqi Li, Hui En Pang, Os- car Qian, Yukun Wei, Zhiqian Lin, Xuanke Shi, Kewang Deng, Xiaoyang Han, Zukai Chen, Xiangyu Fan, Han- ming Deng, ... on PitharXiv
  15. [35] Rui Yang, Ziyu Zhu, Yanwei Li, Jingjia Huang, Shen Yan, Siyuan Zhou, Zhe Liu, Xiangtai Li, Shuangye Li, Wenqian Wang, Yi Lin, and Hengshuang Zhao. Visual Spatial Tuning. arXiv preprint arXiv:2511.05491, 2025 on PitharXiv
  16. [36] Shusheng Yang, Jihan Yang, Pinzhi Huang, Ellis Brown, Zi- hao Yang, Yue Yu, Shengbang Tong, Zihan Zheng, Yifan Xu, Muhan Wang, Daohan Lu, Rob Fergus, Yann LeCun, Li Fei- Fei, and Saining Xie. Cambrian-S: Towards Spatial Super- sensing in Video.arXiv preprint arXiv:2511.04670, 2025 on PithPith's reviewarXiv
  17. [37] Kun Ouyang, Yuanxin Liu, Haoning Wu, Yi Liu, Hao Zhou, Jie Zhou, Fandong Meng, and Xu Sun. SpaceR: Reinforc- ing MLLMs in Video Spatial Reasoning.arXiv preprint arXiv:2504.01805, 2025 on PithPith's reviewarXiv
  18. [38] Junfei Wu, Jian Guan, Kaituo Feng, Qiang Liu, Shu Wu, Liang Wang, Wei Wu, and Tieniu Tan. Reinforc- ing Spatial Reasoning in Vision-Language Models with In- terwoven Thinking and Visual Drawing.arXiv preprint arXiv:2506.09965, 2025 on PithPith's reviewarXiv
  19. [39] Haiwen Diao, Jiahao Wang, Penghao Wu, Yuhao Dong, Yuwei Niu, Yue Zhu, Zhongang Cai, Weichen Fan, Lin- 10 jun Dai, Silei Wu, Xuanyu Zheng, Mingxuan Li, Yuanhan Zhang, Bo Li, Hanming Deng, Huchuan Lu, Quan Wang, Lei Yang, Lewei Lu, Dahua Lin, and Ziwei Liu. From Pix- els to Word... on PithPith's reviewarXiv
  20. [40] Zhanpeng Luo, Ce Zhang, Silong Yong, Cunxi Dai, Qian- wei Wang, Haoxi Ran, Guanya Shi, Katia Sycara, and Yaqi Xie. pySpatial: Generating 3D Visual Programs for Zero- Shot Spatial Reasoning.arXiv preprint arXiv:2603.00905, 2026 on PitharXiv
  21. [41] Jiahui Zhang, Yurui Chen, Yanpeng Zhou, Yueming Xu, Ze Huang, Jilin Mei, Junhui Chen, Yu-Jie Yuan, Xinyue Cai, Guowei Huang, Xingyue Quan, Hang Xu, and Li Zhang. From Flatland to Space: Teaching Vision-Language Models to Perceive and Reason in 3D.arXiv preprint arXiv:2503.22976, 2025 on PitharXiv
  22. [42] Sihan Yang, Runsen Xu, Yiman Xie, Sizhe Yang, Mo Li, Jingli Lin, Chenming Zhu, Xiaochen Chen, Haodong Duan, Xiangyu Yue, Dahua Lin, Tai Wang, and Jiangmiao Pang. MMSI-Bench: A Benchmark for Multi-Image Spatial Intel- ligence.arXiv preprint arXiv:2505.23764, 2025 on PithPith's reviewarXiv
  23. [43] Jihan Yang, Shusheng Yang, Anjali W. Gupta, Rilyn Han, Li Fei-Fei, and Saining Xie. Thinking in Space: How Multi- modal Large Language Models See, Remember, and Recall Spaces.arXiv preprint arXiv:2412.14171, 2024 on PithPith's reviewarXiv
  24. [44] Xingyu Fu, Yushi Hu, Bangzheng Li, Yu Feng, Haoyu Wang, Xudong Lin, Dan Roth, Noah A. Smith, Wei-Chiu Ma, and Ranjay Krishna. BLINK: Multimodal Large Lan- guage Models Can See but Not Perceive.arXiv preprint arXiv:2404.12390, 2024 on PithPith's reviewarXiv
  25. [45] DeepSeek AI. DeepSeek Harness. Software repository, on Pith
  26. [46] OpenAI. Codex CLI. Software repository, 2026. Accessed November 14, 2026. 11 Appendix Contents A Qualitative cases and evidence chains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .13 A.1 Capabilities enabled by complete evidence ch... on Pith
  27. [48] The selected-view run also starts at 3, so acquisition alone does not settle the answer. 3 doors Error link: candidate -> category membership Use selected views to check membership Distinguish architectural doorways from equipment-storage cabinet panels. Review excludes the eq... on Pith
  28. [49] The selected-view run also starts at 3, so acquisition alone does not settle the answer. 3 doors Error link: candidate -> category membership Use selected views to check membership Distinguish architectural doorways from equipment-storage cabinet panels. Review excludes the eq... on Pith
  29. [2026] Accessed November 14, 2026 on Pith

Challenges and discussion

Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.

No comments yet.

Sign in to comment