STAIRS-Former: Spatio-Temporal Attention with Interleaved Recursive Structure Transformer for Offline Multi-task Multi-agent Reinforcement Learning

• URL: http://arxiv.org/abs/2603.11691v1
• Date: Thu, 12 Mar 2026 08:56:20 GMT
• Title: STAIRS-Former: Spatio-Temporal Attention with Interleaved Recursive Structure Transformer for Offline Multi-task Multi-agent Reinforcement Learning
• Abstract summary: STAIRS-Former is a transformer architecture augmented with spatial and temporal hierarchies.<n>We show that STAIRS-Former consistently outperforms prior methods and achieves new state-of-the-art performance.
• Score: 19.174135233947958
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Offline multi-agent reinforcement learning (MARL) with multi-task datasets is challenging due to varying numbers of agents across tasks and the need to generalize to unseen scenarios. Prior works employ transformers with observation tokenization and hierarchical skill learning to address these issues. However, they underutilize the transformer attention mechanism for inter-agent coordination and rely on a single history token, which limits their ability to capture long-horizon temporal dependencies in partially observable MARL settings. In this paper, we propose STAIRS-Former, a transformer architecture augmented with spatial and temporal hierarchies that enables effective attention over critical tokens while capturing long interaction histories. We further introduce token dropout to enhance robustness and generalization across varying agent populations. Extensive experiments on diverse multi-agent benchmarks, including SMAC, SMAC-v2, MPE, and MaMuJoCo, with multi-task datasets demonstrate that STAIRS-Former consistently outperforms prior methods and achieves new state-of-the-art performance.

Related papers

• OneRank: Unified Transformer-Native Ranking Architecture for Multi-Task Recommendation [28.67407481470947]
  Multi-task learning is essential in recommender systems to enable complementary learning among diverse user feedback.<n>OneRank is a Transformer-centric multi-task ranking framework that eliminates encoder-predictorprivate separation.<n>OneRank significantly outperforms state-of-the-art baselines in offline and online experiments on large-scale industrial datasets.
  arXiv  Detail & Related papers  (2026-06-15T15:16:17Z)
• MARTI-MARS$^2$: Scaling Multi-Agent Self-Search via Reinforcement Learning for Code Generation [64.2621682259008]
  Multi-Agent Reinforced Training and Inference Framework with Self-Search Scaling (MARTI-MARS2)<n>We propose a Multi-Agent Reinforced Training and Inference Framework with Self-Search Scaling (MARTI-MARS2) to integrate policy learning with multi-agent tree search.<n>We show that MARTI-MARS2 achieves 77.7%, outperforming strong baselines like GPT-5.1 on challenging code generation benchmarks.
  arXiv  Detail & Related papers  (2026-02-08T07:28:44Z)
• Diffusion Forcing for Multi-Agent Interaction Sequence Modeling [52.769202433667125]
  MAGNet is a unified autoregressive diffusion framework for multi-agent motion generation.<n>It supports a wide range of interaction tasks through flexible conditioning and sampling.<n>It captures both tightly synchronized activities and loosely structured social interactions.
  arXiv  Detail & Related papers  (2025-12-19T18:59:02Z)
• ToolACE-MT: Non-Autoregressive Generation for Agentic Multi-Turn Interaction [84.90394416593624]
  Agentic task-solving with Large Language Models (LLMs) requires multi-turn, multi-step interactions.<n>Existing simulation-based data generation methods rely heavily on costly autoregressive interactions between multiple agents.<n>We propose a novel Non-Autoregressive Iterative Generation framework, called ToolACE-MT, for constructing high-quality multi-turn agentic dialogues.
  arXiv  Detail & Related papers  (2025-08-18T07:38:23Z)
• Skip-Layer Attention: Bridging Abstract and Detailed Dependencies in Transformers [56.264673865476986]
  This paper introduces Skip-Layer Attention (SLA) to enhance Transformer models. SLA improves the model's ability to capture dependencies between high-level abstract features and low-level details. Our implementation extends the Transformer's functionality by enabling queries in a given layer to interact with keys and values from both the current layer and one preceding layer.
  arXiv  Detail & Related papers  (2024-06-17T07:24:38Z)
• An Effective-Efficient Approach for Dense Multi-Label Action Detection [23.100602876056165]
  It is necessary to simultaneously learn (i) temporal dependencies and (ii) co-occurrence action relationships. Recent approaches model temporal information by extracting multi-scale features through hierarchical transformer-based networks. We argue that combining this with multiple sub-sampling processes in hierarchical designs can lead to further loss of positional information.
  arXiv  Detail & Related papers  (2024-06-10T11:33:34Z)
• Multi-layer Learnable Attention Mask for Multimodal Tasks [2.378535917357144]
  Learnable Attention Mask (LAM) strategically designed to globally regulate attention maps and prioritize critical tokens. LAM adeptly captures associations between tokens in BERT-like transformer network. Comprehensive experimental validation on various datasets, such as MADv2, QVHighlights, ImageNet 1K, and MSRVTT.
  arXiv  Detail & Related papers  (2024-06-04T20:28:02Z)
• Transformer RGBT Tracking with Spatio-Temporal Multimodal Tokens [13.608089918718797]
  We propose a novel Transformer-T tracking approach, which mixes multimodal tokens from static templates and multimodal search Transformer to handle target appearance changes. Our module is inserted into the transformer network and inherits joint feature extraction, searchtemplate matching, and cross-temporal interaction. Experiments on three RGBT benchmark datasets show that the proposed approach maintains competitive performance compared to other state-of-the-art tracking algorithms.
  arXiv  Detail & Related papers  (2024-01-03T11:16:38Z)
• PAT: Position-Aware Transformer for Dense Multi-Label Action Detection [36.39340228621982]
  We present PAT, a transformer-based network that learns complex temporal co-occurrence action dependencies in a video. We embed relative positional encoding in the self-attention mechanism and exploit multi-scale temporal relationships. We evaluate the performance of our proposed approach on two challenging dense multi-label benchmark datasets.
  arXiv  Detail & Related papers  (2023-08-09T16:29:31Z)
• FormerTime: Hierarchical Multi-Scale Representations for Multivariate Time Series Classification [53.55504611255664]
  FormerTime is a hierarchical representation model for improving the classification capacity for the multivariate time series classification task. It exhibits three aspects of merits: (1) learning hierarchical multi-scale representations from time series data, (2) inheriting the strength of both transformers and convolutional networks, and (3) tacking the efficiency challenges incurred by the self-attention mechanism.
  arXiv  Detail & Related papers  (2023-02-20T07:46:14Z)
• UPDeT: Universal Multi-agent Reinforcement Learning via Policy Decoupling with Transformers [108.92194081987967]
  We make the first attempt to explore a universal multi-agent reinforcement learning pipeline, designing one single architecture to fit tasks. Unlike previous RNN-based models, we utilize a transformer-based model to generate a flexible policy. The proposed model, named as Universal Policy Decoupling Transformer (UPDeT), further relaxes the action restriction and makes the multi-agent task's decision process more explainable.
  arXiv  Detail & Related papers  (2021-01-20T07:24:24Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.