Techniques for Adaptive Robotics

When building autonomous systems we often had to adapt the changes in context over time,  advances in the techniques needed to do this efficiently are always interesting.   Note here the use of trial and error for real time adaptation.   Interesting, but s that efficient or even risky depending on the tasks involved?

Robots Are Being Programmed to Adapt in Real Time 
Horizon: The EU Research & Innovation Magazine
By Gareth Willmer

Jean-Baptiste Mouret of the National Institute for Research in Computer Science and Automation in France, as part of the European ResiBots initiative, is designing a lower-cost robot that can function for long periods without the need for human maintenance, and is better at overcoming unexpected obstacles. Mouret's ResiBots team is using micro-data-learning algorithms, which can help robots adapt on the fly in a similar way to how animals respond to problems. In contrast, most conventional robots self-diagnose a problem before working out a solution. The new method involves having the robots learn what alternative actions they can take proactively by trial and error, which could help them overcome difficulties and prevent shutdown in situations such as disaster scenarios. Said Mouret, ‘We’re not trying to solve everything. I’m more interested in how they can adapt — and, in fact, adapting to what’s happening is some of what makes animals intelligent.” ... '