Reward Functions for Accelerated Learning

By Maja J. Mataric. 1994

How researchers see this paper

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

Citations in the corpus

Reinforcement learning in robotics: A survey (2013, background): “nell (1992); Mataric (1997); Nemec et al (2009, 2010); Oßwald et al (2010); Paletta et al (2007); Platt et al (2006); Riedmiller et al (2009);”
Autonomous Agents and Multi-agent Systems Manuscript No. Learning Potential Functions and Their Representations for Multi-task Reinforcement Learning (unknown): “In RL, it may refer either to training the agent on successive tasks of increasing complexity, until the desired complexity is reached [17,27,57,59,61,65], or, more commonly, to supplementing the MDP’s reward function with additional, artificial rewards [3,14,15,24,38,49,50,53,85].”
Reward Shaping via Meta-Learning (2019, unknown): “The technique ﬁrst emerges in animal training (Skinner, 1990), and is then introduced to RL (Dorigo & Colombetti, 1994; Mataric, 1994) to tackle increasingly complex problems like Doom (Wu & Tian, 2017) and Dota 2 (OpenAI, 2018).”
Imitation Learning with Demonstrations and Shaping Rewards (2014, background): “Mataric 1994; Randlov and Alstrom 1998), and can be viewed as rules-of-thumb that provide short-term information about the goodness/badness of states and actions.”
Purposive behavior acquisition for a real robot by vision-based reinforcement learning (1995, background): “However, this is not always the case in the real world, where everything changes asynchronously [9].”
Learning Organizational Roles in a Heterogeneous Multi-agent System (1995, background): “Mataric(Mataric, 1994) discusses the concept of progress estimators akin to the idea of potential.”
Transfer Learning via Inter-Task Mappings for Temporal Difference Learning (2007, methodology): “Q-value Reuse may be considered a type of reward shaping (Colombetti and Dorigo, 1993; Mataric, 1994): we are able to directly use the expected rewards from the source task to bias the learner in the target task.”
Principled Methods for Advising Reinforcement Learning Agents (2003, background): “In order to mitigate " thrashing " – repeatedly trying ineffective actions – rewards should be supplied as often as possible (Mataric, 1994).”
Embodied Evolution: Distributing an evolutionary algorithm in a population of robots (2002, background): “Thus, the ETP approach appears to provide a feasible way forward for ER.”
Reinforcement Learning and Markov Decision Processes (2012, background): “Another way to help the agent is by shaping [Mataric(1994), Dorigo and Colombetti(1997), Ng et al(1999)Ng, Harada, and Russell].”
Combining manual feedback with subsequent MDP reward signals for reinforcement learning (2010, result): “Shaping rewards [4, 11] is changing the output of the reward function to learn the same task, as we did in Method 1.”
Transfer via inter-task mappings in policy search reinforcement learning (2007, background): “Reward shaping [2, 5] implicitly guides agents by modifying the reward function.”
Approximate Policy Iteration with a Policy Language Bias: Solving Relational Markov Decision Processes (2011, background): “This might include providing an initial non-trivial policy, providing a heuristic function, or some form of reward shaping (Mataric, 1994).”