A robot that reinforcement-learns to identify and memorize important previous observations

By Bram Bakker, Viktor Zhumatiy, Gabriel Gruener, J. Urgen Schmidhuber. 2003

73 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computational Theory and Mathematics, 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 Artificial Intelligence

In Computational Theory and Mathematics

In Reinforcement Learning

Citations in the corpus

End-to-End Training of Deep Visuomotor Policies (2015, methodology): “small, simple networks (Pomerleau, 1989; Hunt et al., 1992; Bekey and Goldberg, 1992; Lewis et al., 1998; Bakker et al., 2003; Mayer et al., 2006), and has largely been supplanted by methods that use carefully designed policies that can be learned efficiently with reinforcement learning (Kober et al”
Deep learning in neural networks: An overview (2014, methodology): “For example, deep LSTM RNNs were used in this way for RL robots (Bakker et al., 2003).”
Reinforcement learning in robotics: A survey (2013, background): “Model-Based Bakker et al. (2006); Hester et al.”
Control of Memory, Active Perception, and Action in Minecraft (2016, methodology): “The tasks we introduce are inspired by the T-maze experiments (Bakker et al., 2003) as well as Maze-Base (Sukhbaatar et al., 2015a), which has natural language descriptions of mazes available to the agent.”
On Learning to Think: Algorithmic Information Theory for Novel Combinations of Reinforcement Learning Controllers and Recurrent Neural World Models (2015, methodology): “2) was used in this way for RL robots [12].”
Discovering Gated Recurrent Neural Network Architectures (2018, methodology): “…LSTM networks have been used to achieve strong results in the supervised sequence learning problems such as in speech recognition (Graves and Jaitly ((2014)) and machine translation Bahdanau et al. (2015a), their success in POMDP tasks has been limited (Bayer et al. (2009a), Bakker et al. (2003)).”
Fast deep reinforcement learning using online adjustments from the past (2018, background): “There has been a lot of recent work on using memory-augmented neural networks as a function approximation for RL agents: using LSTMs [Bakker et al., 2003, Hausknecht and Stone, 2015], or more sophisticated architectures [Graves et al., 2016, Oh et al., 2016, Wayne et al., 2018].”
Quasi-online reinforcement learning for robots (2006, methodology): “Possible further improvements to the methods explored in this paper include more sophisticated generalization techniques [9], [8], [1], [6], techniques for dealing with partial observability [2], [1], and techniques for easily adding more background knowledge [9].”
Contextual Behaviors and Internal Representations Acquired by Reinforcement Learning with a Recurrent Neural Network in a Continuous State and Action Space Task (2008, methodology): “Against such a trend, there are some works which tackled to the continuous state space problems[4][5][6].”
Reinforcement learning by backpropagation through an LSTM model/critic (2007, unknown): “Our example problem is a variation of an artificial “navigation problem” which we as well as others studied before [10], [1], [2].”
Evolving Deep LSTM-based Memory Networks using an Information Maximization Objective (2016, methodology): “While LSTM networks have been used to achieve strong results in the supervised sequence learning problems such as in speech recognition [10] and machine translation [2], their success in POMDP tasks has been limited [5, 4].”