Competitive reinforcement learning in continuous control tasks

Abramson, Myriam; Pachowicz, Peter W.; Wechsler, Harry

doi:10.1109/ijcnn.2003.1223699

Cited by 7 publications

(7 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The optimal number of steps indicated in (Smart and Kaelbling, 2000) is 56 steps so 60 steps per episode can be considered optimal due to the continuing exploration caused by the -1 reward on every step. Therefore the converged learning result is also as good as in (Abramson et al, 2003) and (Schaefer et al, 2007). (Abramson et al, 2003) used Learning Vector Quantization together with Sarsa instead of using NRBF so the complexity of the learning algorithm is similar to the one used in this paper.…”

Section: Mountain-carmentioning

confidence: 64%

“…Therefore the converged learning result is also as good as in (Abramson et al, 2003) and (Schaefer et al, 2007). (Abramson et al, 2003) used Learning Vector Quantization together with Sarsa instead of using NRBF so the complexity of the learning algorithm is similar to the one used in this paper. The reward function provided more guidance towards the goal than the one used here and a fixed starting point was used instead of a random one.…”

Section: Mountain-carmentioning

confidence: 64%

“…Convergence to an average of 60 steps per episode is achieved after about 40 episodes. This is better than the 'less than 100 episodes' indicated in (Abramson et al, 2003). The optimal number of steps indicated in (Smart and Kaelbling, 2000) is 56 steps so 60 steps per episode can be considered optimal due to the continuing exploration caused by the -1 reward on every step.…”

Section: Mountain-carmentioning

confidence: 95%

See 2 more Smart Citations

Light-Weight Reinforcement Learning With Function Approximation for Real-Life Control Tasks

Främling

2008

Proceedings of the Fifth International Conference on Informatics in Control, Automation and Robotics Service

View full text Add to dashboard Cite

Abstract:Despite the impressive achievements of reinforcement learning (RL) in playing Backgammon already in the beginning of the 90's, relatively few successful real-world applications of RL have been reported since then. This could be due to the tendency of RL research to focus on discrete Markov Decision Processes that make it difficult to handle tasks with continuous-valued features. Another reason could be a tendency to develop continuously more complex mathematical RL models that are difficult to implement and operate. Both of these issues are addressed in this paper by using the gradient-descent Sarsa(λ) method together with a Normalised Radial Basis Function neural net. The experimental results on three typical benchmark control tasks show that these methods outperform most previously reported results on these tasks, while remaining computationally feasible to implement even as embedded software. Therefore the presented results can serve as a reference both regarding learning performance and computational applicability of RL for real-life applications.

show abstract

Section: Mountain-carmentioning

confidence: 64%

Section: Mountain-carmentioning

confidence: 64%

Section: Mountain-carmentioning

confidence: 95%

See 1 more Smart Citation

Light-Weight Reinforcement Learning With Function Approximation for Real-Life Control Tasks

Främling

2008

Proceedings of the Fifth International Conference on Informatics in Control, Automation and Robotics Service

View full text Add to dashboard Cite

show abstract

“…When we deal with an off-policy, the evaluation of the move made in a state will affect the policy from the current move but the move itself is independent of the policy update and therefore we can use a different policy to make the move. On-line policies depend a lot on exploration in order for the action values to be accurate [23].…”

Section: On-policy Vs Off-policy Methodsmentioning

confidence: 99%

Simulated Annealing Based Hierarchical Q-Routing: A Dynamic Routing Protocol

Lopez¹,

Heisterkamp

2011

2011 Eighth International Conference on Information Technology: New Generations

View full text Add to dashboard Cite

“…This function approximation approach separates learning the action value function from learning the state representation (but see [8] for a combined approach). The intermediate reward r c is obtained from the user decision at the choice point while the discounted terminal rewards upon reaching the goal states are obtained from the underlying temporal MDP.…”

Section: Reinforcement Learning Of User Preferencesmentioning

confidence: 99%

Training Coordination Proxy Agents

Abramson

Chao

Mittu

The 2006 IEEE International Joint Conference on Neural Network Proceedings

Self Cite

View full text Add to dashboard Cite

Abstract-Delegating the coordination role to proxy agents can improve the overall outcome of the task at the expense of cognitive overload due to switching subtasks. Stability and commitment are characteristics of human teamwork but must not prevent the detection of better opportunities. In addition, coordination proxy agents must be trained from examples as a single agent but must interact with multiple agents. We apply machine learning techniques to the task of learning team preferences from mixed-initiative interactions and compare the outcome results of different simulated user patterns. This paper introduces a novel approach for the adjustable autonomy of coordination proxies based on the reinforcement learning of abstract actions.

show abstract

Competitive reinforcement learning in continuous control tasks

Cited by 7 publications

References 11 publications

Light-Weight Reinforcement Learning With Function Approximation for Real-Life Control Tasks

Light-Weight Reinforcement Learning With Function Approximation for Real-Life Control Tasks

Simulated Annealing Based Hierarchical Q-Routing: A Dynamic Routing Protocol

Training Coordination Proxy Agents

Contact Info

Product

Resources

About