SPARC: Spatial-Aware Path Planning via Attentive Agent Communication

Sayang Mu; Xiangyu Wu; Bo An

cs.ROMar 3, 2026

SPARC: Spatial-Aware Path Planning via Attentive Agent Communication

Authors: Sayang Mu, Xiangyu Wu, Bo An

Organizations: Nanyang Technological University Singapore

Abstract

Efficient communication is critical for decentralized Multi-Robot Path Planning (MRPP), yet existing learned communication methods treat all neighboring robots equally regardless of their spatial proximity, leading to diluted attention in congested regions where coordination matters most. We propose Relation enhanced Multi Head Attention (RMHA), a communication mechanism that explicitly embeds pairwise Manhattan distances into the attention weight computation, enabling each robot to dynamically prioritize messages from spatially relevant neighbors. Combined with a distance-constrained attention mask and GRU gated message fusion, RMHA integrates seamlessly with MAPPO for stable end-to-end training. In zero-shot generalization from 8 training robots to 128 test robots on 40x40 grids, RMHA achieves approximately 75 percent success rate at 30 percent obstacle density outperforming the best baseline by over 25 percentage points. Ablation studies confirm that distance-relation encoding is the key contributor to success rate improvement in high-density environments. Index Terms-Multi-robot path planning, graph attention mechanism, multi-head attention, communication optimization, cooperative decision-making

Figures & tables

Figure 1. Illustration of MAPF constraints: vertex collisions, edge collisions, and swap collisions among agents must all be avoided. Three diagrams illustrating MAPF collision types that must be avoided: two agents occupying the same vertex (vertex collision), two agents traversing the same edge in opposite directions (edge collision), and two adjacent agents swapping positions (swap collision).

Method	Year	Move	Wait a	Wait b	Coll.	Ego-goal	All-goal	Block	Oscil.	Deviate
PRIMAL	2019	$\downarrow$	$\downarrow$	$\rightarrow$	$\downarrow$	–	$\uparrow$	$\downarrow$	–	–
GNN	2020	–	–	–	–	–	–	–	–	–
MAPPER	2020	$\downarrow$	$\downarrow$	$\downarrow$	$\downarrow$	$\uparrow$	–	–	$\downarrow$	$\downarrow$
G2RL	2020	$\downarrow$	–	–	$\downarrow$	–	$\uparrow$	–	–	–
PRIMAL2	2021	$\downarrow$	$\downarrow$	$\rightarrow$	$\downarrow$	$\uparrow$	–	–	–	–
MAGAT	2021	–	–	–	–	–	–	–	–	–

Table 1. Comparison of Reward Designs in RL-based MAPF Methods.

Figure 2. MARL modeling paradigms for MAPF: MG, POMG, and Dec-POMDP. Schematic comparing three multi-agent RL modeling paradigms for MAPF: a Markov Game with full observability, a Partially Observable Markov Game where each agent sees a local view, and a Dec-POMDP that adds a conditional observation distribution.

Action	Reward
Move (up/down/left/right)	$-0.3$
Stay (at goal / off goal)	$0.0$ / $-0.3$
Collision	$-2.0$
Blocking	$-1.0$

Table 2. Reward Settings

Figure 3. Architecture of the rmha communication module. Block diagram of the RMHA communication module: local observations are combined with Manhattan-distance embeddings, fed through query and key projections, masked by the communication radius, and aggregated via multi-head attention to produce updated messages.

Figure 4. Network architecture of rmha integrated with MAPPO. Full network architecture of RMHA combined with MAPPO, showing the convolutional observation encoder, LSTM temporal fusion, distance-aware multi-head attention communication block, and the four output heads producing policy, value, message vector, and a blocking flag.

Parameter	Value
Training agent count	8
Action count	5
Max episode length	256
FOV size	3
World size	$(10,40)$
Obstacle probability	$(0.0,0.5)$

Table 3. Experimental Parameters

Figure 5. A

10\!\times\!10

grid world with 30% obstacle density and 8 agents. Colored squares indicate start positions; colored circles indicate goals; gray circles represent obstacles; the red dashed box shows the

3\!\times\!3

FOV. Example 10-by-10 grid world used for training, with 30 percent obstacle density and 8 agents. Colored squares mark agent start positions, colored circles mark their goals, gray circles mark static obstacles, and a red dashed box outlines the 3-by-3 field of view.

Figure 6. Success rate of rmha under different agent counts. Line plot showing the success rate of RMHA as the number of agents increases from 16 to 128, with the model trained only at 8 agents. Success rate decays gradually but avoids collapse at the largest scale.

Figure 7. Communication ablation: training reward curves. Training reward curves over time for three ablation variants: MAPPO without communication, MAPPO with graph communication, and RMHA. The two communication-enabled variants converge faster and more stably than MAPPO alone.

Panel of Figure 8 — Figure 8. Communication ablation: success rate comparison. Two bar charts comparing success rate at 0, 15, and 30 percent obstacle density. Left: MAPPO without communication versus graph communication; graph communication wins at every density. Right: graph communication versus RMHA; RMHA wins by a larger margin as density increases.

Figure 9. Success rate comparison across algorithms on Warehouse and City/Game maps. Grouped bar chart comparing success rate of ODrM*, SCRIMP, DHC, PICO, and RMHA on Warehouse and City/Game maps at 0, 15, and 30 percent obstacle complexity. RMHA maintains the highest success rate as complexity grows.

Figure 10. Maximum goals reached across algorithms on Warehouse and City/Game maps. Grouped bar chart of maximum goals reached (out of 128) for each baseline and RMHA across Warehouse and City/Game maps. SCRIMP and RMHA hold above 120 even at 30 percent density while PICO and DHC drop sharply.

Figure 11. Global trajectory comparison: rmha (left) vs. MAPPO (right). Side-by-side overhead view of agent trajectories on the same grid. The RMHA panel on the left shows well-organized, non-overlapping paths; the MAPPO panel on the right shows tangled paths and stalled agents.

Figure 12. Deadlock resolution in a narrow corridor. At

T_{1}

, agents A and B enter a conflict; the baseline leads to deadlock by

T_{2}

–

T_{3}

. Under rmha , agent A yields at

T_{3}

, allowing B to pass, and both proceed by

T_{4}

. Four-step storyboard of two agents A and B meeting head-on in a narrow corridor. The baseline leaves them deadlocked at steps T2 and T3, while under RMHA agent A yields sideways at T3 so agent B passes, and both reach their goals by T4.

Explore similar work

May 8, 2026

Learning to Communicate Locally for Large-Scale Multi-Agent Pathfinding

Multi-agent pathfinding (MAPF) is a widely used abstraction for multi-robot trajectory planning problems, where multiple homogeneous agents move simultaneously within a shared environment. Although solving MAPF optimally is NP-hard, scalable and efficient solvers are critical for real-world applications such as logistics and search-and-rescue. To this end, the research community has proposed various decentralized suboptimal MAPF solvers that leverage machine learning. Such methods frame MAPF (from a single agent perspective) as a Dec-POMDP where at each time step an agent has to decide an action based on the local observation and typically solve the problem via reinforcement learning or imitation learning. We follow the same approach but additionally introduce a learnable communication module tailored to enhance cooperation between agents via efficient feature sharing. We present the Local Communication for Multi-agent Pathfinding (LC-MAPF), a generalizable pre-trained model that applies multi-round communication between neighboring agents to exchange information and improve their coordination. Our experiments show that the introduced method outperforms the existing learning-based MAPF solvers, including IL and RL-based approaches, across diverse metrics in a diverse range of (unseen) test scenarios. Remarkably, the introduced communication mechanism does not compromise LC-MAPF's scalability, a common bottleneck for communication-based MAPF solvers.

Valeriy Vyaltsev, Alsu Sagirova, Anton Andreychuk +5

CogAI Lab, Moscow, Russia

Multi-Agent Coordination Robust MAPF Multi-Agent Communication Multi-Agent Path Finding

Ask PDF

May 26, 2026

SCALE-COMM: Shared, Contrastively-Aligned Latent Embeddings for MARL Communication

Emergent communication enables partially observant Autonomous Mobile Robots (AMRs) to coordinate effectively in decentralized multi-agent reinforcement learning (MARL) settings. However, existing approaches often struggle with unstable communication protocols, ungrounded message semantics, and interference between communication learning and policy optimization, leading to degraded coordination over time. We propose SCALE-COMM (Shared, Contrastively-Aligned Latent Embeddings for COMMunication), a self-supervised framework for learning compact, stable, and policy-relevant communication representations. SCALE-COMM decouples communication learning from policy optimization by training low-dimensional latent messages that capture task-relevant planning and traffic information, while enforcing consistency across agents and time. Across standard MARL benchmarks and a realistic warehouse coordination task, SCALE-COMM consistently outperforms existing communication frameworks in both representation quality and task performance. The learned communication space yields improved stability, sample efficiency, and throughput under policy fine-tuning, demonstrating the effectiveness of representation-driven communication for scalable multi-agent coordination.

Mahmoud Abouelyazid, Eman Hammad

Electrical and Computer Engineering Department Texas A&M University, College Station, TX, USA

Multi-Robot Coordination Decentralized MARL Multi-Agent Communication Emergent Communication

Ask PDF

Sep 12, 2026

Decentralized Multi-Robot Task Allocation Under Degraded Communication: A Benchmark of Performance, Reliability, and Computation

Selecting a decentralized Multi-Robot Task Allocation (MRTA) method for embedded deployment on autonomous platforms requires considering more than route performance alone. We benchmark six decentralized MRTA allocators (CBAA, ACBBA, PI, HIPC, DMCHBA, and DGA) in the Collaborative Visit (CV) scenario to characterize tradeoffs among MinMax and MinSum travel, communication robustness and demand, allocation reliability, computational burden, and scale sensitivity. The core study uses 500 paired ten-target instances across 25 ideal and degraded communication conditions spanning Bernoulli loss, Gilbert--Elliott loss, and Rayleigh fading, with additional campaigns examining pre-allocation, execution-integrated computation, and sensitivity to grid size, robot density, and target load. Across the 24 impaired core conditions, DGA and DMCHBA achieved the lowest mean MinMax travel at 24.49 and 24.78 steps, respectively. HIPC narrowly led mean MinSum travel at 66.95 steps, followed by DGA at 67.22, with both methods occupying the top two in every impaired condition. DMCHBA had the lowest publication intensity at 2.08 publications per team step. In ten-target pre-allocation, HIPC and DMCHBA remained viable and stable in every tested condition, while ACBBA, PI, and DGA lost stability or viability as communication degraded. Under ideal delivery, median full-protocol computation

\Cterm

in the primary ten-target comparison ranged from 4.88 ms for DMCHBA to 1.346 s for DGA. Static route quality preserved DGA and DMCHBA as the leading MinMax methods, while DGA led MinSum at three of four target loads and HIPC led at 50 targets. Static and execution-integrated computation rankings diverged as task load increased. The results identify distinct allocator operating regions across route objective, communication behavior, reliability, and computational constraints.

James Lott, Vahraz Honary

Department of Electrical Engineering, University of San Diego, San Diego, CA 92110 USA

Multi-Robot Task Allocation

Ask PDF