Information Technology Reference
In-Depth Information
4 Conclusion and Future Works
A simple solution to coordinate two robots for moving an object to a final goal
position has been presented. The solution learns for different cases and it gives
a more robust and simple solution. We have considered two solutions based on
the reward at the end of the episode. The second approach considers a smaller
set of states, which helps to produce an early converge. As we explain we do
not need to use nine states —the first approach— to resolve our reinforcement
learning problem. We only consider the nine states in case we use more possible
actions that include stop one for the robots.
We are going to advance in two different lines. First, we are going to employ
the Webots simulator [9] in order to get a more realistic results. Then, we will use
real robots in a real environment. We are already developing some procedures
for computing the robots and stick positions by using an external vision system.
Also, we are going to explore other reward/punishment schemes. Particularly
we want to study if a specific reward for the stick orientation could improve the
general performance.
References
1. Martin, J.A., de Lope, J., Maravall, D.: Analysis and solution of a predator-
protector-prey multi-robot system by a high-level reinforcement learning architec-
ture and adaptive systems theory. Neurocomputing 58(12), 1266-1272 (2010)
2. Iima, H., Kuroe, Y.: Swarm Reinforcement Learning Algortithms Based on Sarsa
Method. In: SICE Annual Conference (2008)
3. Yang, E., Gu, D.: Multiagent Reinforcement Learning for Multi-Robot Systems:
A Survey. CSM-404. Technical Reports of the Department of Computer Science,
University of Essex (2004)
4. Matarić, M.J.: Coordination and learning in Multi-Robot Systems, pp. 6-8. IEEE
Computer Society Press, Los Alamitos (1998)
5. Matarić, M.J.: Reinforcement Learning in the Multi-Robot Domain. Autonomous
Robots 4(1), 73-83 (1997)
6. Sutton, R.S., Barto, A.G.: Reinforcement Learning: An Introduction. MIT Press,
Cambridge (1998)
7. Maravall, D., De Lope, J., Martín H, J.A.: Hybridizing evolutionary computation
and reinforcement learning for the design of almost universal controllers for au-
tonomous robots. Neurocomputing 72(4-6), 887-894 (2009)
8. Sutton, R.S.: Reinforcement learning architectures. In: Proc. Int. Symp. on Neural
Information Processing, Kyushu Inst., Kyushu Inst. of Technology, Japan (1992)
9. Webots. Commercial Mobile Robot Simulation Software,
http://www.cyberbotics.com
Search WWH ::
Custom Search