mlsanityData to 7 Oct 2026

Zero-Shot Task Generalization with Multi-Task Deep Reinforcement Learning

arXiv 1706.05064 ↗First posted 15 Jun 2017cs.AI cs.LGcited by 97

Not on the map. Why?

Abstract (arXiv)

As a step towards developing zero-shot task generalization capabilities in reinforcement learning (RL), we introduce a new RL problem where the agent should learn to execute sequences of instructions after learning useful skills that solve subtasks. In this problem, we consider two types of generalizations: to previously unseen instructions and to longer sequences of instructions. For generalization over unseen instructions, we propose a new objective which encourages learning correspondences between similar subtasks by making analogies. For generalization over sequential instructions, we present a hierarchical architecture where a meta controller learns to use the acquired skills for executing the instructions. To deal with delayed reward, we propose a new neural architecture in the meta controller that learns when to update the subtask, which makes learning more efficient. Experimental results on a stochastic 3D domain show that the proposed ideas are crucial for generalization to longer instructions as well as unseen instructions.

Key facts. Cited by 97 papers in the map (data through 7 Oct 2026). First posted on arXiv 15 Jun 2017.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2018: 2; 2019: 0; 2020: 1; 2021: 21; 2022: 22; 2023: 15; 2024: 18; 2025: 9; 2026: 9.

97 papers.

Sort:

By area:

  1. 2022Do As I Can, Not As I Say: Grounding Language in Robotic Affordancescited by 2,302
  2. 2023Describe, Explain, Plan and Select: Interactive Planning with Large Language Models Enables Open-World Multi-Task Agentscited by 396
  3. 2024Agent Workflow Memorycited by 342
  4. 2022Video PreTraining (VPT): Learning to Act by Watching Unlabeled Online Videoscited by 338
  5. 2020Language Conditioned Imitation Learning over Unstructured Datacited by 208
  6. 2023JARVIS-1: Open-World Multi-task Agents with Memory-Augmented Multimodal Language Modelscited by 170
  7. 2021Contrastive Behavioral Similarity Embeddings for Generalization in Reinforcement Learningcited by 123
  8. 2021Learning Language-Conditioned Robot Behavior from Offline Data and Crowd-Sourced Annotationcited by 117
  9. 2023Diffusion Models for Reinforcement Learning: A Surveycited by 84
  10. 2024Optimus-1: Hybrid Multimodal Memory Empowered Agents Excel in Long-Horizon Taskscited by 83
  11. 2018Learning to Understand Goal Specifications by Modelling Rewardcited by 80
  12. 2022Does Zero-Shot Reinforcement Learning Exist?cited by 79
38 more
  1. 2021LTL2Action: Generalizing LTL Instructions for Multi-Task RLcited by 72
  2. 2024HiRT: Enhancing Robotic Control with Hierarchical Robot Transformerscited by 72
  3. 2023MP5: A Multi-modal Open-ended Embodied System in Minecraft via Active Perceptioncited by 67
  4. 2021Learning to Synthesize Programs as Interpretable and Generalizable Policiescited by 59
  5. 2022TASKOGRAPHY: Evaluating robot task planning over large 3D scene graphscited by 59
  6. 2024Multi-Task Learning for Routing Problem with Cross-Problem Zero-Shot Generalizationcited by 49
  7. 2021Grounding Language to Entities and Dynamics for Generalization in Reinforcement Learningcited by 45
  8. 2021Adversarial Skill Chaining for Long-Horizon Robot Manipulation via Terminal State Regularizationcited by 44
  9. 2023Sequential Dexterity: Chaining Dexterous Policies for Long-Horizon Manipulationcited by 43
  10. 2023Open-World Multi-Task Control Through Goal-Aware Representation Learning and Adaptive Horizon Predictioncited by 39
  11. 2022Environment Generation for Zero-Shot Compositional Reinforcement Learningcited by 32
  12. 2023Plan, Eliminate, and Track -- Language Models are Good Teachers for Embodied Agentscited by 32
  13. 2022LISA: Learning Interpretable Skill Abstractions from Languagecited by 30
  14. 2021Pointer Value Retrieval: A new benchmark for understanding the limits of neural network generalizationcited by 29
  15. 2021JueWu-MC: Playing Minecraft with Sample-efficient Hierarchical Reinforcement Learningcited by 28
  16. 2024Odyssey: Empowering Minecraft Agents with Open-World Skillscited by 26
  17. 2022Learning Options via Compressioncited by 22
  18. 2022Intelligent problem-solving as integrated hierarchical reinforcement learningcited by 21
  19. 2021Cross-Trajectory Representation Learning for Zero-Shot Generalization in RLcited by 20
  20. 2024DeepLTL: Learning to Efficiently Satisfy Complex LTL Specifications for Multi-Task RLcited by 20
  21. 2025Pareto Set Learning for Multi-Objective Reinforcement Learningcited by 18
  22. 2021In a Nutshell, the Human Asked for This: Latent Goals for Following Temporal Specificationscited by 17
  23. 2024EXTRACT: Efficient Policy Learning by Extracting Transferable Robot Skills from Offline Datacited by 13
  24. 2025Position: Interactive Generative Video as Next-Generation Game Enginecited by 13
  25. 2021CLASTER: Clustering with Reinforcement Learning for Zero-Shot Action Recognitioncited by 12
  26. 2023Combining Behaviors with the Successor Features Keyboardcited by 12
  27. 2021Generalization in Text-based Games via Hierarchical Reinforcement Learningcited by 11
  28. 2024Reward Machines for Deep RL in Noisy and Uncertain Environmentscited by 11
  29. 2021SILG: The Multi-environment Symbolic Interactive Language Grounding Benchmarkcited by 10
  30. 2022Perceiving the World: Question-guided Reinforcement Learning for Text-based Gamescited by 10
  31. 2022Noisy Symbolic Abstractions for Deep RL: A case study with Reward Machinescited by 10
  32. 2021Multitasking Inhibits Semantic Driftcited by 8
  33. 2023Conceptual Reinforcement Learning for Language-Conditioned Taskscited by 8
  34. 2023Improving Long-Horizon Imitation Through Instruction Predictioncited by 8
  35. 2023Maximum diffusion reinforcement learningcited by 8
  36. 2025One Subgoal at a Time: Zero-Shot Generalization to Arbitrary Linear Temporal Logic Requirements in Multi-Task Reinforcement Learningcited by 8
  37. 2018CompILE: Compositional Imitation Learning and Executioncited by 7
  38. 2023Composing Task Knowledge with Modular Successor Feature Approximatorscited by 7

50 most cited of 97 All 97 (JSON)