Partially Observable Zero-shot coordination by Predicting Intention of Partner

Jinnyeong Yang; Yuhwan Jeong; Hoyong Kwon; Minseok Kim; Jihun Kim; Kuk-Jin Yoon

cs.AIOct 6, 2026

Partially Observable Zero-shot coordination by Predicting Intention of Partner

Authors: Jinnyeong Yang, Yuhwan Jeong, Hoyong Kwon, Minseok Kim, Jihun Kim, Kuk-Jin Yoon

Organizations: KAIST, Visual Intelligence Lab

Abstract

Zero-shot coordination in embodied settings requires acting while the partner is intermittently out of view, leaving existing methods with ambiguous partner representations and uncertainty over hidden partner states. We propose Predicting Intention of Partner (PIP) to jointly address these challenges. PIP uses a Joint-view VAE to distill richer training-time evidence from the union of both agents' local observations into a partner representation available from local observations alone. Partner-state Belief networks further infer the partner's hidden location and behavioral tendencies from the ego agent's interaction history. We evaluate PIP in Burrito-PO, Overcooked-PO, and a Melting Pot substrate, together with a human evaluation in Burrito-PO. PIP attains the highest mean performance among the compared methods across all three benchmarks. Human evaluation and diagnostic analyses further support coordination with unseen partners and the contributions of both components under partner occlusion.

Figures & tables

Figure 1 : Illustration of coordination under dynamic and spatial partial observability in Burrito-PO. (a) Under full observation, the partner state is directly available. (b) Under a restricted local view, the partner may leave the visible region, creating ambiguity over its location and behavior.

Method	Open	Hallway	FC	Ring
FCP	141.6 $\pm$ 9.6	57.1 $\pm$ 2.7	21.9 $\pm$ 6.0	48.3 $\pm$ 6.1
MEP	115.9 $\pm$ 10.3	58.2 $\pm$ 2.6	17.6 $\pm$ 5.8	51.1 $\pm$ 5.4
E3T	55.3 $\pm$ 3.6	44.5 $\pm$ 5.7	21.0 $\pm$ 4.1	27.5 $\pm$ 3.1
ERS	101.0 $\pm$ 7.8	36.6 $\pm$ 4.5	24.0 $\pm$ 4.3	48.9 $\pm$ 5.5
GAMMA	163.3 $\pm$ 10.6	57.7 $\pm$ 3.3	19.8 $\pm$ 5.3	48.5 $\pm$ 5.6
GOAT	165.5 $\pm$ 10.6	61.5 $\pm$ 5.1	19.0 $\pm$ 5.3	77.3 $\pm$ 5.9

Table 1: Evaluation with 12 held-out behavior-preference partners unseen during training. The best result in each layout is shown in bold . Mean

\pm

standard error.

Table 3

Figure 2 : Human evaluation in the Burrito-PO environment. Participants interacted with GAMMA, GOAT, and our method. Results report (a) sparse reward, (b) cooperation preference ranking, and (c) post-interaction survey scores on coordination quality.

Appendix figures & tables23 assets

Supplementary material from the paper’s appendix.

Appendix

Preference	Reward Shape
Plating Ingredients	$-30,\,20$
Washing Plates	$-30,\,20$
Delivering Dishes	$-30,\,20$
Chopping Ingredients	$-30,\,20$
Potting Rice	$-30,\,20$
Grilling Meat/Mushroom	$-30,\,20$

Appendix

Table B.1 : Event-based BP features and corresponding reward design in Burrito-PO.

Events	Weights
Put an onion / dish / soup / tomato	0
onto the counter	0
Pickup an onion / dish / soup / tomato	0
from the counter	0
Pickup an onion from the onion dispenser	$-20,0,10$
Pickup an tomato from the tomato dispenser	$-20,0,10$

Appendix

Table B.2: Event-based BP features and corresponding reward design in Overcooked-PO

Figure C.1 : Partner-visibility ratio (%) by view size and layout, measured from Self-Play rollouts using the MEP algorithm.

Hyperparameter	Burrito-PO	Overcooked-PO	Melting Pot
CNN kernels	[3, 3], [3, 3], [3, 3]	[3, 3], [3, 3], [3, 3]	[8, 8], [4, 4], [3, 3]
CNN channels	[32, 64, 32]
hidden layer size	[64]
recurrent layer size	64
activation function	ReLU
weight decay	0

Appendix

Table C.1 : Hyperparameters for policy models in each environment.

Behavioral Belief
Hyperparameter	Value
$\lambda_{\mathrm{behavior}}$	0.1
$K_{\mathrm{off}}$	10
$\gamma_{\mathrm{rec}}$	0.90
$\gamma_{\mathrm{ten}}$	0.97
$\alpha_{\mathrm{rec}}$	1.0

Appendix

Table C.2 : Partner-state Belief hyperparameters for target policy models.

Metric	Baseline	PIP (Ours)	Ratio
Total parameters	463K	591K	1.28 $\times$
Actor rollout inference time	1.13 ms	2.05 ms	1.81 $\times$
Actor rollout FLOPs	0.381G	0.764G	2.01 $\times$

Appendix

Table C.3: Representative computational costs at the training rollout batch size of 64. Baseline corresponds to GAMMA / GOAT, whose target policies share an identical architecture.

shared across stages
hyperparameter	Stage 1 (teacher pretrain)		Stage 2 (student distill)
trained module	teacher (enc + dec)		student (enc + dec), teacher frozen
encoder input	$(o_{1}\cup o_{2})\oplus o_{\text{partner}}$		$o_{\text{partner}}$
$\lambda_{KD}$	–		10.0
epoch	500		500
hyperparameter	Burrito-PO	Overcooked-PO		Melting Pot

Appendix

Table C.4 : Hyperparameters for the Joint-view VAE.

Panel of Figure 14 — Figure E.1 : Burrito environment layouts used in our experiments. Each layout induces different coordination patterns under ego-centric partial observability.

Figure E.2 : The Overcooked environment layout used in our experiments, counter circuit.

Panel of Figure 16 — Figure E.3 : The Melting Pot coop mining maps used in our experiments. Grey walls occlude the agents’ view, iron ore (grey) gives $+1$ when mined alone, whereas gold ore (yellow) gives $+8$ to each agent only if both mine the same ore within a 3-step window.

Figure F.1 : t-SNE visualization of episode-level posterior means from each encoder on held-out trajectories from 12 BP partners (144 episodes per partner; perplexity 50, PCA initialization). Each point corresponds to one episode and is colored by the partner policy that generated it, with the same color used for the same partner across panels. Compared with (a) Fixed

\sigma

and (b) Het

\sigma

, (c) Ours produces more clearly separated partner-specific latent clusters.

Figure F.2 : Partner-state Belief analysis. (a) Current Location Belief accuracy by Time Since Last Seen (TSLS). (b)

\Delta

-step future Location Belief accuracy. (c)

\Delta

-step future Behavioral Belief accuracy.

Method	Score
GAMMA	143.3 $\pm$ 11.2
GOAT	103.0 $\pm$ 8.6
PIP	256.2 $\pm$ 15.1

Appendix

Table F.1: Full-observation coordination performance on the Burrito Open layout with 12 held-out behavior-preference partners.

Method	Cap.	Sup.	Reward
GAMMA	No	No	163.3 $\pm$ 10.6
+ Capacity control	Yes	No	145.8 $\pm$ 5.7
+ Supervision control	No	Yes	88.7 $\pm$ 10.2
+ Joint control	Yes	Yes	96.0 $\pm$ 7.1
PIP (Ours)	Yes	Yes	249.0 $\pm$ 12.2

Appendix

Table F.2: Capacity and supervision controls on Burrito-PO Open. Cap. denotes capacity approximately matched to PIP, and Sup. denotes direct auxiliary supervision using privileged partner-state labels.

Figure F.3 : Return of PIP against the

M{=}12

partners as the corruption rate

r

increases from

0

0.2

(mean

\pm

s.e.), under sparse (blackout) and noisy (per-channel shuffle) observation corruption. The dashed line is GAMMA with clean observations (

r=0

Figure G.1 : Participant consent form for the Burrito human-AI coordination study, outlining the study purpose, procedure, anonymous data collection, exclusion of personal identifiers, data protection policy, potential risks, voluntary participation, and withdrawal conditions.

#	Step	Instruction
1	Movement	Use the arrow keys to move in all four directions. Participants were informed that the game only resumes after pressing Enter or the yellow “Continue” button.
2	Read menu	Inspect the order cards at the bottom of the screen. Each card indicates the requested burrito type and the remaining ticks before the order expires. Orders can be delivered in any sequence.
3	Pick meat	Move to the meat dispenser and press Space to pick up a piece of meat.
4	Chop meat	Place the meat on the cutting board, press Space repeatedly to chop it, and pick up the resulting chopped meat.
5	Grill meat	Place the chopped meat on the grill to cook it.
6	Boil rice	Pick up rice from the rice dispenser and place it in the pot to boil.

Appendix

Table G.1 : Tutorial steps provided before the human-AI coordination survey. The tutorial familiarized participants with the game controls, object interactions, burrito recipes, and failure-recovery mechanics.

Panel of Figure 24 — Figure G.2 : Tutorial and survey interfaces. The left panel shows step-by-step instructions, object icons, and the ego-centric partial-observation view. The right panel shows the main-survey gameplay interface, where participants cooperate with an anonymized AI partner under ego-centric partial observability.

Figure G.3 : Post-game evaluation materials. The left panel shows the post-episode evaluation survey assessing each AI partner after gameplay. The right panel shows the final preference ranking of the three AI partners based on collaboration experience.

Metric	GAMMA	GOAT	PIP (Ours)	vs. GAMMA	vs. GOAT
Sparse reward mean $\pm$ SE	$355.7\pm 19.7$	$370.4\pm 18.4$	$476.0\pm 22.7$	$+33.8\%$	$+28.5\%$
Top-1 preference	$10/66$ , $15.2\%$	$8/66$ , $12.1\%$	$48/66$ , $72.7\%$	$+57.5$ pts.	$+60.6$ pts.

Appendix

Table G.2 : Human evaluation performance and preference summary (

N=66

participants).

Outcome	Friedman test		PIP vs. GAMMA		PIP vs. GOAT
Outcome	$\chi^{2}(2)$	$p$	$p$	$r$	$p$	$r$
Sparse reward	34.95	$2.57{\times}10^{-8}$	$2.43{\times}10^{-7}$	$+0.74$	$3.16{\times}10^{-6}$	$+0.66$
Preference rank	37.85	$6.04{\times}10^{-9}$	$4.80{\times}10^{-6}$	$+0.62$	$1.16{\times}10^{-6}$	$+0.67$

Appendix

Table G.3 : Within-subject statistical tests for task performance and preference ranking.

Category	GAMMA	GOAT	PIP	vs. GAMMA	vs. GOAT
Coordination	$3.68$	$3.26$	$5.01$	$+36.1\%$	$+53.7\%$
Predictability	$3.33$	$3.11$	$4.83$	$+45.0\%$	$+55.3\%$
Adaptability	$3.29$	$3.11$	$4.89$	$+48.6\%$	$+57.2\%$
Role Complementarity	$3.09$	$2.80$	$4.35$	$+40.8\%$	$+55.4\%$
Subjective Experience	$3.33$	$3.13$	$5.02$	$+50.8\%$	$+60.4\%$
Overall	$3.45$	$3.18$	$5.27$	$+52.8\%$	$+65.7\%$

Appendix

Table G.4 : Category-level survey scores and relative improvements.

Category	Friedman test		PIP vs. GAMMA		PIP vs. GOAT
Category	$\chi^{2}(2)$	$p$	$p$	$r$	$p$	$r$
Coordination	35.46	$1.99{\times}10^{-8}$	$2.14{\times}10^{-5}$	$+0.62$	$1.42{\times}10^{-7}$	$+0.76$
Predictability	39.07	$3.28{\times}10^{-9}$	$6.64{\times}10^{-7}$	$+0.72$	$4.31{\times}10^{-8}$	$+0.82$
Adaptability	33.95	$4.25{\times}10^{-8}$	$1.14{\times}10^{-5}$	$+0.66$	$1.23{\times}10^{-6}$	$+0.72$
Role Complementarity	28.31	$7.12{\times}10^{-7}$	$7.10{\times}10^{-5}$	$+0.60$	$3.01{\times}10^{-6}$	$+0.73$
Subjective Experience	32.57	$8.47{\times}10^{-8}$	$1.20{\times}10^{-6}$	$+0.73$	$5.86{\times}10^{-7}$	$+0.76$

Appendix

Table G.5 : Supporting within-subject statistical tests for post-episode survey scores. Wilcoxon

p

-values are uncorrected, and all PIP-vs-baseline comparisons remain significant after Holm–Bonferroni correction.

Explore similar work

Grounded Joint-Attention Other-Play for Zero-Shot Coordination
Oct 5, 2026Giulia Benintendi, Constantin Ruhdorfer, Fabian Kögel +1Multi-Agent Coordination
University of Zurich · University of Stuttgart
Shaping Zero-Shot Coordination via State Blocking
May 12, 2026Mingu Kang, Sunwoo Lee, Yonghyeon Jo +1Multi-Agent Coordination Coordination
Graduate School of Artificial Intelligence UNIST Ulsan, South Korea 44919
Beyond Partner Diversity: An Influence-Based Team Steering Framework for Zero-Shot Human-Machine Teaming
May 14, 2026Wei Sheng, Rohan PalejaHuman-Ai Collaboration Artificial Intelligence Agents
Department of Computer Science Purdue University

Oct 5, 2026, cs.AIJ/K move · Enter open · S save

Grounded Joint-Attention Other-Play for Zero-Shot Coordination

Giulia Benintendi, Constantin Ruhdorfer, Fabian Kögel, Andreas Bulling

University of Zurich · University of Stuttgart

Multi-Agent Coordination

Open paper AskPDF arXiv

Joint attention - the human ability to share a common visual or cognitive focus with others - enables a meeting of minds that lets us coordinate even with unfamiliar partners. In this work we investigate whether equipping AI agents with a similar mechanism can enable such zero-shot coordination. We introduce Mutual Attention for zero-shot TEaming (MATE): a novel multi-agent reinforcement learning method inspired by human joint attention. MATE encourages agents to coordinate their actions by aligning their visual attention on scene-salient objects during the interaction rather than relying on arbitrary partner-dependent conventions established during training. Unlike symmetry-breaking approaches that merely prevent brittle conventions from emerging, MATE actively promotes coordination through an environment-grounded signal that is naturally shared across partners. We evaluate MATE on three benchmarks: our Card Alignment Game, designed to isolate brittle convention formation, and the more challenging Level-Based Foraging and OvercookedV2 benchmarks. Our experiments consistently show that a joint-attention-inspired signal improves coordination with unknown partners, underlining MATE's potential as a general coordination mechanism that complements and surpasses symmetry-breaking approaches.

Abstract

Explore similar work

Grounded Joint-Attention Other-Play for Zero-Shot Coordination

Shaping Zero-Shot Coordination via State Blocking

Beyond Partner Diversity: An Influence-Based Team Steering Framework for Zero-Shot Human-Machine Teaming