• Proceedings of the IEEK Conference
• /
• 2001.06c
• /
• pp.13-16
• /
• 2001

In reinforcement teaming, Q-learning converges quite slowly to a good policy. Its because searching for the goal state takes very long time in a large stochastic domain. So I propose the speedup method using the Q-value initialization for model-free reinforcement learning. In the speedup method, it learns a naive model of a domain and makes boundaries around the goal state. By using these boundaries, it assigns the initial Q-values to the state-action pairs and does Q-learning with the initial Q-values. The initial Q-values guide the agent to the goal state in the early states of learning, so that Q-teaming updates Q-values efficiently. Therefore it saves exploration time to search for the goal state and has better performance than Q-learning. 1 present Speedup Q-learning algorithm to implement the speedup method. This algorithm is evaluated. in a grid-world domain and compared to Q-teaming.

• Journal of KIISE:Software and Applications
• /
• v.31 no.11
• /
• pp.1431-1438
• /
• 2004

Reinforcement learning learns policies for accomplishing a task's goal by experience through interaction between agent and environment. Q-learning, basis algorithm of reinforcement learning, has the problem of curse of dimensionality and slow learning speed in the incipient stage of learning. In order to solve the problems of Q-learning, new function approximation methods suitable for reinforcement learning should be studied. In this paper, to improve these problems, we suggest Fuzzy Q-Map algorithm that is based on online fuzzy clustering. Fuzzy Q-Map is a function approximation method suitable to reinforcement learning that can do on-line teaming and express uncertainty of environment. We made an experiment on the mountain car problem with fuzzy Q-Map, and its results show that learning speed is accelerated in the incipient stage of learning.

• Journal of Korea Multimedia Society
• /
• v.11 no.3
• /
• pp.353-363
• /
• 2008

A multi-agent system has a longer learning period and larger state-spaces than a sin91e agent system. In this paper, we suggest a new method to reduce the learning time of Nash Q-learning in a multi-agent environment. We apply Macro-actions to Nash Q-learning to improve the teaming speed. In the Nash Q-teaming scheme, when agents select actions, rewards are accumulated like Macro-actions. In the experiments, we compare Nash Q-learning using Macro-actions with general Nash Q-learning. First, we observed how many times the agents achieve their goals. The results of this experiment show that agents using Nash Q-learning and 4 Macro-actions have 9.46% better performance than Nash Q-learning using only 4 primitive actions. Second, when agents use Macro-actions, Q-values are accumulated 2.6 times more. Finally, agents using Macro-actions select less actions about 44%. As a result, agents select fewer actions and Macro-actions improve the Q-value's update. It the agents' learning speeds improve.

• The KIPS Transactions:PartB
• /
• v.9B no.2
• /
• pp.155-162
• /
• 2002

Reinforcement learning is a learning method that uses trial-and-error to perform Learning by interacting with dynamic environments. It is classified into online reinforcement learning and delayed reinforcement learning. In this paper, we propose an online reinforcement learning system (ONRELS : Outline REinforcement Learning System). ONRELS updates the estimate-value about all the selectable (state, action) pairs before making state-transition at the current state. The ONRELS learns by interacting with the compressed environments through trial-and-error after it compresses the state space of the mage environments. Through experiments, we can see that ONRELS can search the shortest path faster than Q-learning using TD-ewor and $Q(\lambda{)}$-learning using $TD(\lambda{)}$ in the maze environments.

• Journal of KIISE:Software and Applications
• /
• v.30 no.7_8
• /
• pp.672-680
• /
• 2003

A shopbot is a software agent whose goal is to maximize buyer´s satisfaction through automatically gathering the price and quality information of goods as well as the services from on-line sellers. In the response to shopbots´ activities, sellers on the Internet need the agents called pricebots that can help them maximize their own profits. In this paper we adopts Q-learning, one of the model-free reinforcement learning methods as a price-setting algorithm of pricebots. A Q-learned agent increases profitability and eliminates the cyclic price wars when compared with the agents using the myoptimal (myopically optimal) pricing strategy Q-teaming needs to select a sequence of state-action fairs for the convergence of Q-teaming. When the uniform random method in selecting state-action pairs is used, the number of accesses to the Q-tables to obtain the optimal Q-values is quite large. Therefore, it is not appropriate for universal on-line learning in a real world environment. This phenomenon occurs because the uniform random selection reflects the uncertainty of exploitation for the optimal policy. In this paper, we propose a Mixed Nonstationary Policy (MNP), which consists of both the auxiliary Markov process and the original Markov process. MNP tries to keep balance of exploration and exploitation in reinforcement learning. Our experiment results show that the Q-learning agent using MNP converges to the optimal Q-values about 2.6 time faster than the uniform random selection on the average.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.15 no.4
• /
• pp.406-411
• /
• 2005

In this paper, we present the area-based Q-learning to search a target object using multiple robot. To search the target in Markovian space, the robots should recognize their surrounding at where they are located and generate some rules to act upon by themselves. Under area-based Q-learning, a robot, first of all, obtains 6-distances from itself to environment by infrared sensor which are hexagonally allocated around itself. Second, it calculates 6-areas with those distances then take an action, i.e., turn and move toward where the widest space will be guaranteed. After the action is taken, the value of Q will be updated by relative formula at the state. We set up an experimental environment with five small mobile robots, obstacles, and a target object, and tried to search for a target object while navigating in a unknown hallway where some obstacles were placed. In the end of this paper, we presents the results of three algorithms - a random search, area-based action making (ABAM), and hexagonal area-based Q-teaming.

• The KIPS Transactions:PartB
• /
• v.11B no.2
• /
• pp.207-212
• /
• 2004

This paper presents a reinforcement learning framework for stock trading systems. Trading system parameters are optimized by Q-learning algorithm and neural networks are adopted for value approximation. In this framework, cooperative multiple agents are used to efficiently integrate global trend prediction and local trading strategy for obtaining better trading performance. Agents Communicate With Others Sharing training episodes and learned policies, while keeping the overall scheme of conventional Q-learning. Experimental results on KOSPI 200 show that a trading system based on the proposed framework outperforms the market average and makes appreciable profits. Furthermore, in view of risk management, the system is superior to a system trained by supervised learning.

• 제어로봇시스템학회:학술대회논문집
• /
• 2000.10a
• /
• pp.489-489
• /
• 2000

Recently, Reinforcement Learning(RL) methods have been used far teaming problems in Partially Observable Markov Decision Process(POMDP) environments. Conventional RL-methods, however, have limited applicability to POMDP To overcome the partial observability, several algorithms were proposed [5], [7]. The aim of this paper is to extend our previous algorithm for POMDP, called Labeling Q-learning(LQ-learning), which reinforces incomplete information of perception with labeling. Namely, in the LQ-learning, the agent percepts the current states by pair of observation and its label, and the agent can distinguish states, which look as same, more exactly. Labeling is carried out by a hash-like function, which we call Labeling Function(LF). Numerous labeling functions can be considered, but in this paper, we will introduce several labeling functions based on only 2 or 3 immediate past sequential observations. We introduce the basic idea of LQ-learning briefly, apply it to maze problems, simple POMDP environments, and show its availability with empirical results, look better than conventional RL algorithms.

• International Journal of Control, Automation, and Systems
• /
• v.1 no.2
• /
• pp.236-242
• /
• 2003

This paper presents a cooperation strategy between a human operator and autonomous robots for an interactive robot soccer game, The interactive robot soccer game has been developed to allow humans to join into the game dynamically and reinforce entertainment characteristics. In order to make these games more interesting, a cooperation strategy between humans and autonomous robots on a team is very important. Strategies can be pre-programmed or learned by robots themselves with learning or evolving algorithms. Since the robot soccer system is hard to model and its environment changes dynamically, it is very difficult to pre-program cooperation strategies between robot agents. Q-learning - one of the most representative reinforcement learning methods - is shown to be effective for solving problems dynamically without explicit knowledge of the system. Therefore, in our research, a Q-learning based learning method has been utilized. Prior to utilizing Q-teaming, state variables describing the game situation and actions' sets of robots have been defined. After the learning process, the human operator could play the game more easily. To evaluate the usefulness of the proposed strategy, some simulations and games have been carried out.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.16 no.1
• /
• pp.79-85
• /
• 2006

This paper presents a strategy and tactic for robot soccer using furry logic mediator that determines robot action depending on the positions and the roles of adjacent two robots. Conventional Q-learning algorithm, where the number of states increases exponentially with the number of robots, is not suitable for a robot soccer system, because it needs so much calculation that processing cannot be accomplished in real time. A modular Q-teaming algorithm reduces a number of states by partitioning the concerned area, where mediator algorithm for cooperation of robots is used additionally. The proposed scheme implements the mediator algorithm among robots by fuzzy logic system, where simple fuzzy rules make the calculation easy and hence proper for robot soccer system. The simulation of MiroSot shows the feasibility of the proposed scheme.