Model-Based Reinforcement Learning with an Approximate, Learned Model

By Leonid Kuvayev, Richard S. Sutton. 1996

Subfields: Reinforcement Learning

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Reinforcement Learning

1. Sergey Levine
2. Pieter Abbeel
3. Michael H. Bowling
4. Doina Precup
5. Peter Stone
6. R´emi Munos
7. Marc G. Bellemare
8. Yuxi Li

Citations in the corpus

• Deep Imitative Models for Flexible Inference, Planning, and Control (2018, methodology): “In our experiments, our approach substantially outperforms both MBRL and IL: it can efficiently learn near-perfect driving through the static-world CARLA simulator from just 7,000 trajectories obtained from 19 hours of driving.”
• Dyna-Style Planning with Linear Function Approximation and Prioritized Sweeping (2008, methodology): “The original papers on the Dyna architecture and most subsequent extensions (e.g., Singh 1992; Peng & Williams 1993; Moore & Atkeson 1993; Kuvayev & Sutton 1996) assumed a Markov environment with a tabular representation of states.”
• Reinforcement Learning: An Introduction (1998, background): “Among the few works to consider planning with learned, approximate models are those by Kuvayev and Sutton (1996), Sutton, Szepesvari, Geramifard, and Bowling (2008), Nouri and Littman (2009), and Hester and Stone (2012).”
• Online Model Learning Algorithms for Actor-Critic Control (2015, methodology): “However, instead of using the process model to generate simulated experiences as most model learning RL algorithms do (Kuvayev and Sutton, 1996; Moore and Atkeson, 1993; Sutton, 1992), it uses the model to directly calculate an accurate policy gradient, which accelerates learning compared to other p”