TR2025-015

Generalized Policy Improvement Algorithms with Theoretically Supported Sample Reuse

    •  Queeney, J., Paschalidis, I.C., Cassandras, C.G., "Generalized Policy Improvement Algorithms with Theoretically Supported Sample Reuse", IEEE Transactions on Automatic Control, DOI: 10.1109/​TAC.2024.3454011, Vol. 70, No. 2, pp. 1236-1243, February 2025.
      BibTeX TR2025-015 PDF
      • @article{Queeney2025feb,
      • author = {Queeney, James and Paschalidis, Ioannis Ch. and Cassandras, Christos G.}},
      • title = {Generalized Policy Improvement Algorithms with Theoretically Supported Sample Reuse},
      • journal = {IEEE Transactions on Automatic Control},
      • year = 2025,
      • volume = 70,
      • number = 2,
      • pages = {1236--1243},
      • month = feb,
      • doi = {10.1109/TAC.2024.3454011},
      • url = {https://www.merl.com/publications/TR2025-015}
      • }
  • MERL Contact:
  • Research Areas:

    Control, Dynamical Systems, Robotics

Abstract:

We develop a new class of model-free deep reinforcement learning algorithms for data-driven, learning-based control. Our Generalized Policy Improvement algorithms combine the pol- icy improvement guarantees of on-policy methods with the efficiency of sample reuse, addressing a trade-off between two important deployment requirements for real-world control: (i) practical performance guarantees and (ii) data efficiency. We demonstrate the benefits of this new class of algorithms through extensive experimental analysis on a broad range of simulated control tasks.

 

  • Related Publication

  •  Queeney, J., Paschalidis, I.C., Cassandras, C.G., "Generalized Policy Improvement Algorithms with Theoretically Supported Sample Reuse", arXiv, October 2024.
    BibTeX arXiv
    • @article{Queeney2024oct,
    • author = {Queeney, James and Paschalidis, Ioannis Ch. and Cassandras, Christos G.}},
    • title = {Generalized Policy Improvement Algorithms with Theoretically Supported Sample Reuse},
    • journal = {arXiv},
    • year = 2024,
    • month = oct,
    • url = {https://arxiv.org/abs/2206.13714}
    • }