MDP Playground: A Design and Debug Testbed for Reinforcement Learning
@inproceedings{Rajan2019MDPPA,
title={MDP Playground: A Design and Debug Testbed for Reinforcement Learning},
author={Raghunandan Rajan and Jessica Lizeth Borja Diaz and Suresh Guttikonda and F{\'a}bio Ferreira and Andr{\'e} Biedenkapp and Jan Ole von Hartz and Frank Hutter},
year={2019}
}We present MDP Playground, an efficient testbed for Reinforcement Learning 1 (RL) agents with orthogonal dimensions that can be controlled independently 2 to challenge agents in different ways and obtain varying degrees of hardness in 3 generated environments. We consider and allow control over a wide variety of 4 dimensions, including delayed rewards, rewardable sequences, density of rewards, 5 stochasticity, image representations, irrelevant features, time unit, action range 6 and more. We…
Figures from this paper
figure 1 
figure 2 
figure 3 
figure 4 
figure 5 
figure 6 
figure 7 
figure 8 
figure 9 
figure 10 
figure 11 
figure 12 
figure 13 
figure 14 
figure 15 
figure 16 
figure 17 
figure 18 
figure 19 
figure 20 
figure 21 
figure 22 
figure 23 
figure 24 
figure 25 
figure 26 
figure 27 
figure 28 
figure 29 
figure 30 
figure 31 
figure 32 
figure 33 
figure 34 
figure 35 
figure 36 
figure 37 
figure 38 
figure 39 
figure 40 
figure 41 
figure 42 
figure 43 
figure 44 
figure 45 
figure 46 
figure 47 
figure 48 
figure 49 
figure 50 
figure 51 
figure 52 
figure 53 
figure 54 
figure 55 
figure 56 
figure 57 
figure 58 
figure 59 
figure 60 
figure 61 
figure 62 
figure 63 
figure 64 
figure 65 
figure 66 
figure 67 
figure 68 
figure 69 
figure 70 
figure 71 
figure 72 
figure 73 
figure 74 
figure 75 
figure 76 
figure 77 
figure 82 
figure 86 
figure 89 
figure 90 
figure 96 
figure 99 
figure 100 
figure 106 
figure 114 
figure 115 
figure 116 
figure 117
4 Citations
A Survey of Zero-shot Generalisation in Deep Reinforcement Learning
- Computer ScienceJ. Artif. Intell. Res.
- 2023
It is argued that taking a purely procedural content generation approach to benchmark design is not conducive to progress in ZSG, and it is suggested fast online adaptation and tackling RL-specific problems as some areas for future work on methods for ZSG.
A Survey of Generalisation in Deep Reinforcement Learning
- Computer ScienceArXiv
- 2021
It is argued that taking a purely procedural content generation approach to benchmark design is not conducive to progress in generalisation, and fast online adaptation and tackling RL-specific problems as some areas for future work on methods for generalisation are suggested.
Hardness in Markov Decision Processes: Theory and Practice
- Computer ScienceNeurIPS
- 2022
A systematic survey of the theory of hardness is presented, and four agents in non-tabular versions of Colosseum environments are benchmarked, obtaining results that demonstrate the generality of tabular hardness measures.
The Sandbox Environment for Generalizable Agent Research (SEGAR)
- Computer ScienceArXiv
- 2022
The Sandbox Environment for Generalizable Agent Research (SEGAR) improves the ease and accountability of generalization research in RL, as generalization objectives can be easy designed by specifying task distributions, which in turns allows the researcher to measure the nature of the generalization objective.
129 References
Learning Invariant Feature Hierarchies
- Computer ScienceECCV Workshops
- 2012
A number of unsupervised learning algorithms to train computer vision models that are weakly inspired by the visual cortex will be presented, based on the sparse auto-encoder concept and the effectiveness of these algorithms for learning invariant feature hierarchies will be demonstrated.
565 M
- A. Riedmiller, A. Fidjeland, G. Ostrovski, S. Petersen, C. Beattie, A. Sadik, I. Antonoglou, 566 H. King, D. Kumaran, D. Wierstra, S. Legg, and D. Hassabis. Human-level control through 567 deep reinforcement learning. Nature, 518(7540):529–533
- 2015
533 M
- A. Riedmiller, A. Fidjeland, G. Ostrovski, S. Petersen, C. Beattie, A. Sadik, I. Antonoglou, 534 H. King, D. Kumaran, D. Wierstra, S. Legg, and D. Hassabis. Human-level control through 535 deep reinforcement learning. Nature, 518(7540):529–533
- 2015
H
- van Hasselt, T. Schaul, G. Ostrovski, W. Dabney, D. Horgan, B. Piot, 479 M. Azar, and D. Silver. Rainbow: Combining improvements in deep reinforcement learning. 480 In S. A. McIlraith and K. Q. Weinberger, editors, Proceedings of the Conference on Artificial 481 Intelligence (AAAI’18), pages 3215–32
- 2018
The Arcade Learning Environment: An Evaluation Platform for General Agents (Extended Abstract)
- Computer ScienceIJCAI
- 2013
The promise of ALE is illustrated by developing and benchmarking domain-independent agents designed using well-established AI techniques for both reinforcement learning and planning, and an evaluation methodology made possible by ALE is proposed.
and Sergey Levine
- Soft Actor-Critic: Off-policy maximum 469 entropy deep reinforcement learning with a stochastic actor. In J. G. Dy and A. Krause, editors, 470 Proceedings of the 35th International Conference on Machine Learning (ICML’18), pages 471 1856–1865. PMLR
- 2018
The arcade learning environment: An 437 evaluation platform for general agents
- Journal of Artificial Intelligence Research,
- 2013
H
- van Hoof, and D. Meger. Addressing function approximation error in actor-critic 460 methods. In J. G. Dy and A. Krause, editors, Proceedings of the 35th International Conference 461 on Machine Learning (ICML’18), pages 1582–1591. PMLR
- 2018
and 561 K
- Kavukcuoglu. Asynchronous methods for deep reinforcement learning. In M. Balcan 562 and K. Weinberger, editors, Proceedings of the 33rd International Conference on Machine 563 Learning (ICML’16), volume 48, pages 1928–1937
- 2016
and 529 K
- Kavukcuoglu. Asynchronous methods for deep reinforcement learning. In M. Balcan 530 and K. Weinberger, editors, Proceedings of the 33rd International Conference on Machine 531 Learning (ICML’16), volume 48, pages 1928–1937
- 2016