PRE2018 4 Group1: Difference between revisions
No edit summary |
No edit summary |
||
| Line 29: | Line 29: | ||
==Brainstorm== | ==Brainstorm== | ||
==Subject== | ==Subject== | ||
Researching and redesigning robotic substitutes for service dogs that help blind people navigate | Researching and redesigning robotic substitutes for service dogs that help blind people navigate | ||
==State of the Art research== | |||
===Robots that can adapt like animals=== | |||
====Relevance==== | |||
When designing a robotic service dog, a problem that can occur will be that the dog can become damaged while the user is out with the service dog. The following article is about how a robot can cope with problems in a similar way animals do. | |||
====Summary==== | |||
In this article, a trial and error algorithm is proposed so that the robots can adapt to damage is under two minutes in a similar manner as animals. | |||
Current recovery typically involves two phases. The robot first needs to diagnose itself, which is followed by selecting the best plan to fix the problem. Problem is that it could be that not every situation is foreseen by the designer, and the robot does not have the right diagnosis or contingency plan to cope with the inflicted damage. Animals deal with injuries in a more trail and error based way. A similar algorithm could be implemented in robots to learn the robot different behaviors to injuries without the limitations to the engineers possible damage scenario’s. | |||
The current state-of-the-art algorithms for this are not suitable since they can’t cope with the curse of dimensionality. Other algorithms take up about 15 minutes and need some human demonstrations of some kind. Animals can do it in 2 minutes, so for robots it would be more practical to do so in a similar time. | |||
The main difference between animals and robots is that animals know the search space of behaviors and can therefore adapt intelligently. Robots would need to do the same in order to achieve a similar behavior. Robots used in the article store knowledge in a behavior-performance space. This helps them to cope with injuries by quickly discovering the behavior that would help in the injury at hand. | |||
[https://www.nature.com/articles/nature14422] | |||
==References== | ==References== | ||
[https://en.wikipedia.org/wiki/Help:Cheatsheet] Legged robots- an overview https://journals.sagepub.com/doi/pdf/10.1177/0142331207075610 | [https://en.wikipedia.org/wiki/Help:Cheatsheet] Legged robots- an overview https://journals.sagepub.com/doi/pdf/10.1177/0142331207075610 | ||
Robots that can adapt like animals https://www.nature.com/articles/nature14422 | |||
Revision as of 11:45, 30 April 2019
Group 1
| Group members | Student number | Study |
|---|---|---|
| Lotte van Gessel | 1237708 | l.s.v.gessel@student.tue.nl |
| Piers da Camino Ancona Lopez Soligo | 1015467 | dit mail adres |
| Sander Poot | 1017804 | s.a.poot@student.tue.nl |
| Timon Heuwekemeijer | 1003212 | t.m.heukemeijer@student.tue.nl |
| Jan van Leeuwen | 1261401 | j.a.v.leeuwen@student.tue.nl |
Brainstorm
Subject
Researching and redesigning robotic substitutes for service dogs that help blind people navigate
State of the Art research
Robots that can adapt like animals
Relevance
When designing a robotic service dog, a problem that can occur will be that the dog can become damaged while the user is out with the service dog. The following article is about how a robot can cope with problems in a similar way animals do.
Summary
In this article, a trial and error algorithm is proposed so that the robots can adapt to damage is under two minutes in a similar manner as animals. Current recovery typically involves two phases. The robot first needs to diagnose itself, which is followed by selecting the best plan to fix the problem. Problem is that it could be that not every situation is foreseen by the designer, and the robot does not have the right diagnosis or contingency plan to cope with the inflicted damage. Animals deal with injuries in a more trail and error based way. A similar algorithm could be implemented in robots to learn the robot different behaviors to injuries without the limitations to the engineers possible damage scenario’s. The current state-of-the-art algorithms for this are not suitable since they can’t cope with the curse of dimensionality. Other algorithms take up about 15 minutes and need some human demonstrations of some kind. Animals can do it in 2 minutes, so for robots it would be more practical to do so in a similar time. The main difference between animals and robots is that animals know the search space of behaviors and can therefore adapt intelligently. Robots would need to do the same in order to achieve a similar behavior. Robots used in the article store knowledge in a behavior-performance space. This helps them to cope with injuries by quickly discovering the behavior that would help in the injury at hand. [1]
References
[2] Legged robots- an overview https://journals.sagepub.com/doi/pdf/10.1177/0142331207075610 Robots that can adapt like animals https://www.nature.com/articles/nature14422