pith. sign in

arxiv: 2605.22814 · v1 · pith:4N4DNWGHnew · submitted 2026-05-21 · 💻 cs.LG

Remember to be Curious: Episodic Context and Persistent Worlds for 3D Exploration

Lily Goli , Justin Kerr , Daniele Reda , Alec Jacobson , Andrea Tagliasacchi , Angjoo Kanazawa This is my paper

Pith reviewed 2026-05-22 06:40 UTC · model grok-4.3

classification 💻 cs.LG
keywords curiosity-driven explorationpersistent world modelsepisodic context3D navigationonline 3D reconstructionreinforcement learningindoor environments
0
0 comments X

The pith

Persistent 3D reconstruction and RGB sequence modeling enable effective curiosity-driven exploration in 3D environments.

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

In complex 3D environments, curiosity-driven agents often loop in place because they lack a stable record of the space and their path through it. This paper establishes that a continuously updated online 3D map serves as a persistent world model while a policy that processes sequences of RGB images maintains the necessary episodic history. With these additions the agent explores more thoroughly when trained solely on curiosity signals from realistic home datasets. The same policy then moves to new environments without further training and adapts rapidly to practical tasks such as picking apples or navigating to a target picture.

Core claim

The authors claim that curiosity requires both spatial persistence through a continuously updated world model and episodic context through trajectory history. They implement the persistent model via online 3D reconstruction and the episodic context via a sequence model over raw RGB frames. This lets the agent explore effectively during training and navigate at deployment using only RGB observations.

What carries the argument

Online 3D reconstruction as persistent world model together with sequence model over RGB observations for episodic context in the policy.

If this is right

  • The curiosity-trained agent outperforms RL active mapping baselines on HM3D.
  • It generalizes zero-shot to Gibson and procedurally generated worlds.
  • It adapts efficiently to tasks like apple picking and image-goal navigation.
  • Adaptation outperforms training those tasks from scratch.

Where Pith is reading between the lines

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

  • This combination of persistence and memory may address similar looping issues in other long-horizon RL problems.
  • The method suggests that explicit mapping can complement learned policies in navigation without sacrificing end-to-end RGB control at test time.
  • Extensions could test the approach with noisy real-world sensor data instead of perfect simulation.

Load-bearing premise

The online 3D reconstruction remains accurate enough over long trajectories in cluttered scenes and the RGB sequence model alone provides sufficient memory to avoid state revisits.

What would settle it

If long-horizon trials in cluttered indoor environments show frequent revisits to the same locations or incomplete coverage despite the 3D model and sequence policy, the necessity of these components for effective curiosity would be called into question.

Figures

Figures reproduced from arXiv: 2605.22814 by Alec Jacobson, Andrea Tagliasacchi, Angjoo Kanazawa, Daniele Reda, Justin Kerr, Lily Goli.

Figure 1
Figure 1. Figure 1: Remember to be Curious. Test-time trajectories of our curiosity-driven policy trained with episodic context and a persistent world model. Left: Our agent explores an indoor scene purely from an image stream; waypoints show the agent’s views along the path. Right: Our end-to-end design enables fine-tuning to downstream sparse-reward tasks and zero-shot generalization to AI-generated worlds. Note: the model … view at source ↗
Figure 2
Figure 2. Figure 2: Method overview. The agent (left) encodes each RGB frame into a per-frame token by fusing patch embeddings and DINOv2 features via a learnable query. Tokens are processed by causal temporal self-attention and a linear-attention module with a global hidden state, before an actor-critic head emits the next action and value estimate. The environment (right) executes the action and returns the next observation… view at source ↗
Figure 3
Figure 3. Figure 3: Our agent explores the scenes more thoroughly leading to a higher completeness in 3D scene coverage [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Ablations on the memory capacity of the forward model and the policy network on HM3D [12], show that a persistent forward model is necessary for effective exploration. Further, the agents equipped with memory capacity can achieve higher 3D scene coverage with less concentrated local loops in its trajectory. different input modalities to their respective RL policy module: ANS predicts a top-down map from RG… view at source ↗
Figure 5
Figure 5. Figure 5: After a few fine-tuning episodes on the apple-picking reward, our exploration agent achieves higher [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Left: Our exploration agent, when fine-tuned for a few episodes on image-goal navigation reward, outperforms an agent trained from scratch on this reward alone, redirecting its general-purpose exploration toward a targeted objective. Right: Our agent generalizes zero-shot to AI-generated worlds of different aesthetic and representation. 4 Related Work We survey the most related works here; a more comprehen… view at source ↗
read the original abstract

Exploration is a prerequisite for learning useful behaviors in sparse-reward, long-horizon tasks, particularly within 3D environments. Curiosity-driven reinforcement learning addresses this via intrinsic rewards derived from the mismatch between the agent's predictive model of the world and reality. However, translating this intrinsic motivation to complex, photorealistic environments remains difficult, as agents can become trapped in local loops and receive fresh rewards for revisiting forgotten states. In this work, we demonstrate that this failure stems from a lack of spatial persistence and episodic context. We show that effective curiosity requires a model of the world that is persistent and continuously updated, paired with an agent that maintains an episodic trajectory history to navigate toward novel regions. We achieve this using an online 3D reconstruction as a persistent model of the world, while the agent policy is parameterized as a sequence model over RGB observations to maintain episodic context. This design enables effective exploration during training while allowing the agent to navigate using solely RGB frames at deployment. Trained purely via curiosity on HM3D, our agent outperforms RL-based active mapping baselines and generalizes zero-shot to Gibson and AI-generated worlds. Our end-to-end policy enables efficient adaptation to downstream tasks, such as apple picking and image-goal navigation, outperforming from-scratch baselines. Please see video results at https://recuriosity.github.io/.

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

2 major / 2 minor

Summary. The manuscript presents a curiosity-driven exploration approach for photorealistic 3D environments. It claims that effective intrinsic motivation requires a persistent world model implemented via online 3D reconstruction paired with episodic context supplied by a sequence model over raw RGB observations. The policy is trained end-to-end using only curiosity rewards on HM3D, after which it is reported to outperform RL-based active mapping baselines, to generalize zero-shot to Gibson and procedurally generated worlds, and to enable efficient fine-tuning on downstream tasks such as apple picking and image-goal navigation.

Significance. If the empirical results hold, the work would be a meaningful contribution to embodied exploration and intrinsic motivation. By supplying an explicit spatial substrate for the curiosity signal and a lightweight episodic memory via sequence modeling, the method directly targets the local-loop failure mode that has limited prior curiosity agents in cluttered indoor scenes. The fact that the final policy operates from RGB alone at deployment time is practically attractive. The combination of reconstruction-based persistence with standard sequence modeling is a clean architectural choice that could be adopted more broadly.

major comments (2)
  1. [§4] §4 (Experimental results): The abstract and main claims assert outperformance over RL-based active mapping baselines together with zero-shot transfer, yet the text provides neither the numerical metrics, baseline implementation details, nor ablation tables needed to evaluate those claims. Without these data the strength of evidence for the central thesis cannot be assessed.
  2. [Reconstruction and reward section] Reconstruction and reward section: The method relies on the online 3D reconstruction remaining sufficiently accurate and complete to generate a reliable curiosity signal across hundreds to thousands of steps in cluttered HM3D scenes. No quantitative reconstruction-quality metrics (coverage, drift, or map-vs-ground-truth overlap) are reported, leaving the load-bearing premise that the persistent map correctly labels novel versus visited regions unverified.
minor comments (2)
  1. [Figure 1] The architecture diagram would benefit from explicit arrows showing how the reconstruction output is used only for the intrinsic reward and not as input to the policy at deployment time.
  2. [§3.2] A short paragraph clarifying the exact form of the episodic context (length of RGB history, attention mechanism, etc.) would improve reproducibility.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive assessment of the work's potential contribution to embodied exploration and for the constructive comments. We address each major point below and will revise the manuscript accordingly to strengthen the empirical support.

read point-by-point responses

  1. Referee: [§4] §4 (Experimental results): The abstract and main claims assert outperformance over RL-based active mapping baselines together with zero-shot transfer, yet the text provides neither the numerical metrics, baseline implementation details, nor ablation tables needed to evaluate those claims. Without these data the strength of evidence for the central thesis cannot be assessed.

    Authors: We agree that the current presentation of results in Section 4 lacks sufficient detail for independent evaluation. In the revised manuscript we will add tables reporting all numerical metrics (success rates, coverage, and efficiency) for the main comparisons, explicit implementation details for the RL-based active mapping baselines (including hyperparameters and training protocols), and additional ablation tables isolating the contributions of persistent reconstruction and episodic sequence modeling. These changes will directly support the claims of outperformance and zero-shot transfer to Gibson and synthetic environments. revision: yes

  2. Referee: [Reconstruction and reward section] Reconstruction and reward section: The method relies on the online 3D reconstruction remaining sufficiently accurate and complete to generate a reliable curiosity signal across hundreds to thousands of steps in cluttered HM3D scenes. No quantitative reconstruction-quality metrics (coverage, drift, or map-vs-ground-truth overlap) are reported, leaving the load-bearing premise that the persistent map correctly labels novel versus visited regions unverified.

    Authors: We acknowledge that direct quantitative validation of reconstruction quality is important for confirming the reliability of the curiosity signal. We will add metrics on reconstruction coverage, drift over long trajectories, and overlap with ground-truth maps (where simulator access permits) to the revised manuscript. While the end-to-end exploration performance on HM3D provides indirect support for the map's utility, these explicit measurements will better substantiate the premise that the persistent 3D model correctly distinguishes novel from visited regions. revision: yes

Circularity Check

0 steps flagged

No circularity detected in derivation chain

full rationale

The paper presents an empirical approach to curiosity-driven exploration using an online 3D reconstruction as a persistent world model and a sequence model over RGB frames for episodic context. Performance claims (outperformance on HM3D, zero-shot transfer to Gibson) are validated through training and evaluation on external benchmarks rather than any closed mathematical derivation. No equations, fitted parameters renamed as predictions, or self-citation chains reduce the central results to inputs by construction. The method relies on established reconstruction and RL components with results measured against independent baselines.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the domain assumption that online 3D reconstruction yields a usable persistent world model and that RGB sequence modeling suffices for episodic context; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (2)
  • domain assumption Online 3D reconstruction from RGB can serve as a sufficiently accurate and continuously updated persistent model of the environment.
    Invoked when the paper states that the agent uses an online 3D reconstruction as the persistent world model.
  • domain assumption A sequence model over RGB observations is adequate to maintain episodic trajectory history for navigation decisions.
    Stated when the agent policy is parameterized as a sequence model over RGB observations to maintain episodic context.

pith-pipeline@v0.9.0 · 5784 in / 1396 out tokens · 46322 ms · 2026-05-22T06:40:35.389579+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

68 extracted references · 68 canonical work pages · 6 internal anchors

  1. [1]

    Cognitive maps in rats and men.Psychological review, 55(4):189, 1948

    Edward C Tolman. Cognitive maps in rats and men.Psychological review, 55(4):189, 1948. 2

    work page 1948

  2. [2]

    Lim, Abhinav Gupta, Li Fei-Fei, and Ali Farhadi

    Yuke Zhu, Roozbeh Mottaghi, Eric Kolve, Joseph J. Lim, Abhinav Gupta, Li Fei-Fei, and Ali Farhadi. Target-driven visual navigation in indoor scenes using deep reinforcement learning. In ICRA, 2017. 2

    work page 2017

  3. [3]

    DD-PPO: Learning near-perfect pointgoal navigators from 2.5 billion frames

    Erik Wijmans, Abhishek Kadian, Ari Morcos, Stefan Lee, Irfan Essa, Devi Parikh, Manolis Savva, and Dhruv Batra. DD-PPO: Learning near-perfect pointgoal navigators from 2.5 billion frames. InICLR, 2020. 2

    work page 2020

  4. [4]

    A possibility for implementing curiosity and boredom in model-building neural controllers

    Jürgen Schmidhuber. A possibility for implementing curiosity and boredom in model-building neural controllers. InFrom Animals to Animats: Proceedings of the First International Confer- ence on Simulation of Adaptive Behavior. MIT Press/Bradford Books, 1991. 2, 14

    work page 1991

  5. [5]

    Curiosity-driven exploration by self-supervised prediction

    Deepak Pathak, Pulkit Agrawal, Alexei A Efros, and Trevor Darrell. Curiosity-driven exploration by self-supervised prediction. InICML, 2017. 2, 7, 9, 14

    work page 2017

  6. [6]

    Advancing Open-source World Models

    Robbyant Team. Advancing open-source world models.arXiv preprint arXiv:2601.20540,

    work page internal anchor Pith review Pith/arXiv arXiv

  7. [7]

    Learning to explore using active neural slam

    Devendra Singh Chaplot, Dhiraj Gandhi, Saurabh Gupta, Abhinav Gupta, and Ruslan Salakhut- dinov. Learning to explore using active neural slam. InICLR, 2020. 2, 6, 9, 14

    work page 2020

  8. [8]

    Ramakrishnan, Ziad Al-Halah, and Kristen Grauman

    Santhosh K. Ramakrishnan, Ziad Al-Halah, and Kristen Grauman. Occupancy anticipation for efficient exploration and navigation. InECCV, 2020. 6, 14

    work page 2020

  9. [9]

    Gleam: Learning generalizable exploration policy for active mapping in complex 3d indoor scenes

    Xiao Chen, Tai Wang, Quanyi Li, Tao Huang, Jiangmiao Pang, and Tianfan Xue. Gleam: Learning generalizable exploration policy for active mapping in complex 3d indoor scenes. In ICCV, 2025. 2, 6, 9, 14

    work page 2025

  10. [10]

    Stanley, and Jeff Clune

    Adrien Ecoffet, Joost Huizinga, Joel Lehman, Kenneth O. Stanley, and Jeff Clune. First return, then explore. InNature, 2021. 2, 9

    work page 2021

  11. [11]

    Learning exploration policies for navigation

    Tao Chen, Saurabh Gupta, and Abhinav Gupta. Learning exploration policies for navigation. In ICLR, 2019. 2, 9, 14

    work page 2019

  12. [12]

    Habitat-matterport 3d dataset (HM3d): 1000 large-scale 3d environments for embodied AI

    Santhosh Kumar Ramakrishnan, Aaron Gokaslan, Erik Wijmans, Oleksandr Maksymets, Alexan- der Clegg, John M Turner, Eric Undersander, Wojciech Galuba, Andrew Westbury, Angel X Chang, Manolis Savva, Yili Zhao, and Dhruv Batra. Habitat-matterport 3d dataset (HM3d): 1000 large-scale 3d environments for embodied AI. InNeurIPS, 2021. 2, 6, 7, 8

    work page 2021

  13. [13]

    Zamir, Zhiyang He, Alexander Sax, Jitendra Malik, and Silvio Savarese

    Fei Xia, Amir R. Zamir, Zhiyang He, Alexander Sax, Jitendra Malik, and Silvio Savarese. Gibson Env: real-world perception for embodied agents. InCVPR, 2018. 2, 6

    work page 2018

  14. [14]

    Generating worlds

    World Labs. Generating worlds. https://www.worldlabs.ai/blog/generating-worlds andhttps://www.worldlabs.ai/blog/spark-2.0, 2024. Accessed: 2025. 2, 8

    work page 2024

  15. [15]

    3d gaussian splatting as markov chain monte carlo

    Shakiba Kheradmand, Daniel Rebain, Gopal Sharma, Weiwei Sun, Yang-Che Tseng, Hossam Isack, Abhishek Kar, Andrea Tagliasacchi, and Kwang Moo Yi. 3d gaussian splatting as markov chain monte carlo. InNeurIPS, 2024. 4, 6, 15 10

    work page 2024

  16. [16]

    Freeman, Joshua B

    Vincent Sitzmann, Semon Rezchikov, William T. Freeman, Joshua B. Tenenbaum, and Fredo Durand. Light field networks: Neural scene representations with single-evaluation rendering. InNeurIPS, 2021. 4

    work page 2021

  17. [17]

    Dinov2: Learning robust visual features without supervision

    Maxime Oquab and Timothée Darcet et al. Dinov2: Learning robust visual features without supervision. InTMLR, 2023. 4

    work page 2023

  18. [18]

    Learning to (learn at test time): Rnns with expressive hidden states

    Yu Sun and Xinhao Li et al. Learning to (learn at test time): Rnns with expressive hidden states. InICML, 2025. 5

    work page 2025

  19. [19]

    LoGeR: Long-Context Geometric Reconstruction with Hybrid Memory

    Junyi Zhang, Charles Herrmann, Junhwa Hur, Chen Sun, Ming-Hsuan Yang, Forrester Cole, Trevor Darrell, and Deqing Sun. Loger: Long-context geometric reconstruction with hybrid memory.arXiv preprint arXiv:2603.03269, 2026. 5, 15

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  20. [20]

    Proximal Policy Optimization Algorithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347, 2017. 5, 15

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  21. [21]

    Sapg: Split and aggregate policy gradients

    Jayesh Singla, Ananye Agarwal, and Deepak Pathak. Sapg: Split and aggregate policy gradients. InICML, 2024. 5

    work page 2024

  22. [22]

    Adam: A method for stochastic optimization

    Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. InICLR,

    work page

  23. [23]

    Habitat 3.0: A co-habitat for humans, avatars and robots,

    Xavi Puig and Eric Undersander et al. Habitat 3.0: A co-habitat for humans, avatars and robots,

    work page

  24. [24]

    What does really matter in image goal navigation? In3DV, 2026

    Gianluca Monaci, Philippe Weinzaepfel, and Christian Wolf. What does really matter in image goal navigation? In3DV, 2026. 6

    work page 2026

  25. [25]

    Active neural mapping

    Zike Yan, Haoxiang Yang, and Hongbin Zha. Active neural mapping. InICCV, 2023. 7, 9, 14

    work page 2023

  26. [26]

    Learning to Navigate in Complex Environments

    Piotr Mirowski, Razvan Pascanu, Fabio Viola, Hubert Soyer, Andrew J Ballard, Andrea Banino, Misha Denil, Ross Goroshin, Laurent Sifre, Koray Kavukcuoglu, et al. Learning to navigate in complex environments.arXiv preprint arXiv:1611.03673, 2016. 8

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  27. [27]

    Exploration by random network distillation

    Yuri Burda, Harrison Edwards, Amos Storkey, and Oleg Klimov. Exploration by random network distillation. InICLR, 2019. 9, 14

    work page 2019

  28. [28]

    Self-supervised exploration via disagree- ment

    Deepak Pathak, Dhiraj Gandhi, and Abhinav Gupta. Self-supervised exploration via disagree- ment. InICML, 2019. 9, 14

    work page 2019

  29. [29]

    Episodic curiosity through reachability

    Nikolay Savinov, Anton Raichuk, Raphaël Marinier, Damien Vincent, Marc Pollefeys, Timothy Lillicrap, and Sylvain Gelly. Episodic curiosity through reachability. InICLR, 2019. 9, 14

    work page 2019

  30. [30]

    Never give up: Learning directed exploration strategies

    Adrià Puigdomènech Badia and Pablo Sprechmann et al. Never give up: Learning directed exploration strategies. InICLR, 2020. 14

    work page 2020

  31. [31]

    Exploration via elliptical episodic bonuses

    Mikael Henaff, Roberta Raileanu, Minqi Jiang, and Tim Rocktäschel. Exploration via elliptical episodic bonuses. InNeurIPS, 2022. 14

    work page 2022

  32. [32]

    Go beyond imagination: maximizing episodic reachability with world models

    Yao Fu, Run Peng, and Honglak Lee. Go beyond imagination: maximizing episodic reachability with world models. InICML, 2023. 9, 14

    work page 2023

  33. [33]

    Improving intrinsic exploration by creating stationary objectives

    Roger Creus Castanyer, Joshua Romoff, and Glen Berseth. Improving intrinsic exploration by creating stationary objectives. InICLR, 2024. 9, 14

    work page 2024

  34. [34]

    Fisherrf: Active view selection and uncertainty quantification for radiance fields using fisher information

    Wen Jiang, Boshu Lei, and Kostas Daniilidis. Fisherrf: Active view selection and uncertainty quantification for radiance fields using fisher information. InECCV, 2024. 9, 14

    work page 2024

  35. [35]

    MAGICIAN: Efficient Long- Term Planning with Imagined Gaussians for Active Mapping

    Shiyao Li, Antoine Guédon, Shizhe Chen, and Vincent Lepetit. MAGICIAN: Efficient Long- Term Planning with Imagined Gaussians for Active Mapping. InCVPR, 2026. 14

    work page 2026

  36. [36]

    Multimodal llm guided explo- ration and active mapping using fisher information

    Wen Jiang, Boshu Lei, Katrina Ashton, and Kostas Daniilidis. Multimodal llm guided explo- ration and active mapping using fisher information. InICCV, 2025. 9, 14 11

    work page 2025

  37. [37]

    Metric-Free Exploration for Topological Mapping by Task and Motion Imitation in Feature Space

    Yuhang He, Irving Fang, Yiming Li, Rushi Bhavesh Shah, and Chen Feng. Metric-Free Exploration for Topological Mapping by Task and Motion Imitation in Feature Space. InRSS,

    work page

  38. [38]

    Ball and Jakob Bauer et al

    Philip J. Ball and Jakob Bauer et al. Genie 3: A new frontier for world models, 2025. URL https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/ . 9, 15

    work page 2025

  39. [39]

    World Action Models are Zero-shot Policies

    Seonghyeon Ye and Yunhao Ge et al. Dreamzero: World action models are zero-shot policies. arXiv preprint arXiv:2602.15922, 2026. 9, 15

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  40. [40]

    Out of sight, out of mind? evaluating state evolution in video world models.arXiv preprint arXiv:2603.13215, 2026

    Ziqi Ma, Mengzhan Liufu, and Georgia Gkioxari. Out of sight, out of mind? evaluating state evolution in video world models.arXiv preprint arXiv:2603.13215, 2026. 9, 15

    work page arXiv 2026

  41. [41]

    3d gaussian splatting for real-time radiance field rendering

    Bernhard Kerbl, Georgios Kopanas, Thomas Leimkühler, and George Drettakis. 3d gaussian splatting for real-time radiance field rendering. InSIGGRAPH, 2023. 9, 15

    work page 2023

  42. [42]

    Splatam: Splat, track & map 3d gaussians for dense rgb-d slam

    Nikhil Keetha, Jay Karhade, Krishna Murthy Jatavallabhula, Gengshan Yang, Sebastian Scherer, Deva Ramanan, and Jonathon Luiten. Splatam: Splat, track & map 3d gaussians for dense rgb-d slam. InCVPR, 2024. 9, 15

    work page 2024

  43. [43]

    State entropy maximization with random encoders for efficient exploration

    Younggyo Seo, Lili Chen, Jinwoo Shin, Honglak Lee, Pieter Abbeel, and Kimin Lee. State entropy maximization with random encoders for efficient exploration. InICML, 2021. 14

    work page 2021

  44. [44]

    Maxin- forl: Boosting exploration in reinforcement learning through information gain maximization

    Bhavya Sukhija, Stelian Coros, Andreas Krause, Pieter Abbeel, and Carmelo Sferrazza. Maxin- forl: Boosting exploration in reinforcement learning through information gain maximization. In ICLR, 2025

    work page 2025

  45. [45]

    Tenenbaum, Tim Rocktäschel, and Edward Grefenstette

    Andres Campero, Roberta Raileanu, Heinrich Küttler, Joshua B. Tenenbaum, Tim Rocktäschel, and Edward Grefenstette. Learning with amigo: Adversarially motivated intrinsic goals. In ICLR, 2021. 14

    work page 2021

  46. [46]

    Yuri Burda, Harri Edwards, Deepak Pathak, Amos Storkey, Trevor Darrell, and Alexei A. Efros. Large-scale study of curiosity-driven learning. InICLR, 2019. 14

    work page 2019

  47. [47]

    Ride: Rewarding impact-driven exploration for procedurally-generated environments

    Roberta Raileanu and Tim Rocktäschel. Ride: Rewarding impact-driven exploration for procedurally-generated environments. InICLR, 2020. 14

    work page 2020

  48. [48]

    Focus on impact: Indoor exploration with intrinsic motivation

    Roberto Bigazzi, Federico Landi, Silvia Cascianelli, Lorenzo Baraldi, Marcella Cornia, and Rita Cucchiara. Focus on impact: Indoor exploration with intrinsic motivation. InRA-L, 2022. 14

    work page 2022

  49. [49]

    Bayes’ Rays: Uncertainty quantification in neural radiance fields

    Lily Goli, Cody Reading, Silvia Sellán, Alec Jacobson, and Andrea Tagliasacchi. Bayes’ Rays: Uncertainty quantification in neural radiance fields. InCVPR, 2024. 14

    work page 2024

  50. [50]

    Uncertainty-driven planner for exploration and navigation

    Georgios Georgakis, Bernadette Bucher, Anton Arapin, Karl Schmeckpeper, Nikolai Matni, and Kostas Daniilidis. Uncertainty-driven planner for exploration and navigation. InICRA, 2022

    work page 2022

  51. [51]

    Macarons: Mapping and coverage anticipation with rgb online self-supervision

    Antoine Guédon, Tom Monnier, Pascal Monasse, and Vincent Lepetit. Macarons: Mapping and coverage anticipation with rgb online self-supervision. InCVPR, 2023

    work page 2023

  52. [52]

    Nextbest- path: Efficient 3d mapping of unseen environments

    Shiyao Li, Antoine Guedon, Clémentin Boittiaux, Shizhe Chen, and Vincent Lepetit. Nextbest- path: Efficient 3d mapping of unseen environments. In Y . Yue, A. Garg, N. Peng, F. Sha, and R. Yu, editors,ICLR, 2025

    work page 2025

  53. [53]

    Activesplat: High-fidelity scene reconstruction through active gaussian splatting

    Yuetao Li, Zijia Kuang, Ting Li, Qun Hao, Zike Yan, Guyue Zhou, and Shaohui Zhang. Activesplat: High-fidelity scene reconstruction through active gaussian splatting. InRA-L, 2025

    work page 2025

  54. [54]

    Activegamer: Active gaussian mapping through efficient rendering

    Liyan Chen, Huangying Zhan, Kevin Chen, Xiangyu Xu, Qingan Yan, Changjiang Cai, and Yi Xu. Activegamer: Active gaussian mapping through efficient rendering. InCVPR, 2025

    work page 2025

  55. [55]

    Rt-guide: Real-time gaussian splatting for information-driven exploration

    Yuezhan Tao, Dexter Ong, Varun Murali, Igor Spasojevic, Pratik Chaudhari, and Vijay Kumar. Rt-guide: Real-time gaussian splatting for information-driven exploration. InRA-L, 2025. 12

    work page 2025

  56. [56]

    Mapex: Indoor structure exploration with probabilistic information gain from global map predictions

    Cherie Ho, Seungchan Kim, Brady Moon, Aditya Parandekar, Narek Harutyunyan, Chen Wang, Katia Sycara, Graeme Best, and Sebastian Scherer. Mapex: Indoor structure exploration with probabilistic information gain from global map predictions. InICRA, 2025

    work page 2025

  57. [57]

    Pipe planner: Pathwise information gain with map predictions for indoor robot exploration

    Seungjae Baek, Brady Moon, Seungchan Kim, Muqing Cao, Cherie Ho, Sebastian Scherer, and Jeong Hwan Jeon. Pipe planner: Pathwise information gain with map predictions for indoor robot exploration. InIROS, 2025. 14

    work page 2025

  58. [58]

    Explore with long-term memory: A benchmark and multimodal llm-based reinforcement learning framework for embodied exploration

    Sen Wang, Bangwei Liu, Zhenkun Gao, Lizhuang Ma, Xuhong Wang, Yuan Xie, and Xin Tan. Explore with long-term memory: A benchmark and multimodal llm-based reinforcement learning framework for embodied exploration. InCVPR, 2026. 14

    work page 2026

  59. [59]

    Where to look next: Learning viewpoint recommendations for informative trajectory planning

    Max Lodel, Bruno Brito, Alvaro Serra-Gómez, Laura Ferranti, Robert Babuška, and Javier Alonso-Mora. Where to look next: Learning viewpoint recommendations for informative trajectory planning. InICRA, 2022. 14

    work page 2022

  60. [60]

    Gennbv: Generalizable next-best-view policy for active 3d reconstruction

    Xiao Chen, Quanyi Li, Tai Wang, Tianfan Xue, and Jiangmiao Pang. Gennbv: Generalizable next-best-view policy for active 3d reconstruction. InCVPR, 2024. 14

    work page 2024

  61. [61]

    On-the-fly reconstruction for large-scale novel view synthesis from unposed images

    Andreas Meuleman, Ishaan Shah, Alexandre Lanvin, Bernhard Kerbl, and George Drettakis. On-the-fly reconstruction for large-scale novel view synthesis from unposed images. InTOG,

    work page

  62. [62]

    Dust3r: Geometric 3d vision made easy

    Shuzhe Wang, Vincent Leroy, Yohann Cabon, Boris Chidlovskii, and Jerome Revaud. Dust3r: Geometric 3d vision made easy. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2024. 15

    work page 2024

  63. [63]

    Efros, and Angjoo Kanazawa

    Qianqian Wang, Yifei Zhang, Aleksander Holynski, Alexei A. Efros, and Angjoo Kanazawa. Continuous 3d perception model with persistent state. InCVPR, 2025. 15

    work page 2025

  64. [64]

    Vggt: Visual geometry grounded transformer

    Jianyuan Wang, Minghao Chen, Nikita Karaev, Andrea Vedaldi, Christian Rupprecht, and David Novotny. Vggt: Visual geometry grounded transformer. InCVPR, 2025. 15

    work page 2025

  65. [65]

    Lyra 2.0: Explorable Generative 3D Worlds

    Tianchang Shen and Sherwin Bahmani et al. Lyra 2.0: Explorable generative 3d worlds.arXiv preprint arXiv:2604.13036, 2026. 15

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  66. [66]

    Learning 3d persistent embodied world models

    Siyuan Zhou, Yilun Du, Yuncong Yang, Lei Han, Peihao Chen, Dit-Yan Yeung, and Chuang Gan. Learning 3d persistent embodied world models. InNeurIPS, 2025

    work page 2025

  67. [67]

    Video world models with long-term spatial memory.arXiv preprint arXiv:2506.05284, 2025

    Tong Wu, Shuai Yang, Ryan Po, Yinghao Xu, Ziwei Liu, Dahua Lin, and Gordon Wetzstein. Video world models with long-term spatial memory.arXiv preprint arXiv:2506.05284, 2025. 15

    work page arXiv 2025

  68. [68]

    High- dimensional continuous control using generalized advantage estimation

    John Schulman, Philipp Moritz, Sergey Levine, Michael Jordan, and Pieter Abbeel. High- dimensional continuous control using generalized advantage estimation. InICLR, 2016. 15 13 A Video Results Please refer to our website,recuriosity.github.io, for video results of our agent exploring diverse 3D environments and performing downstream tasks. B Related Work...

    work page 2016