Researchers at the University of Washington have come up with an approach where robots would be working or learning new skills via collaborating working by shooting a set of query to a larger online community. Rajesh Rao one of the lead scientist asserted that they are taking the one to one interaction to the next stage where the machine would be asking help from the whole world in case it happen to stuck at some problem or a puzzle.
Until now, robots could only acquire learning through imitating humans, which of course guzzled time but this new approach would make things easier and consume less time. However, the basic steps would be ingrained into the machine’s system and then it would be required to ask for additional input from the online community. Idea is to collect more info and finish the task properly with higher efficiency.
Machine-learning techniques require huge amount of data and crowdsourcing is the optimal solution in such a scenario.
Tapping the online crowdsourcing community
The team designed a study to teach robot a model-building task by tapping the online crowdsourcing community. Participants provided basic steps for building few lego models like car, tree, turtle & snake. These ‘basic steps’ were not enough for the machines to develop the complete models. So, they turned to the online world, specifically to a crowdsourcing site called Amazon Mechanical Turk. From the huge database of the crowd-generated models, the robot selected the model that is based on difficulty to construct based on the community’s rating and coincided with the original at hand. Eventually, the robot was able to develop the best model given at its dispose.
This form of learning is termed as the goal-based imitation. Where target is reaching the end product and not the steps or the process. For instance, a robot may see the steps that are involved in building any lego model however, with the intervention of huge online database, it can re-create the model but with different steps all together and with a probability of much simpler and finer steps relatively.
The technology as well as the approach looks promising. Robots would be able to leverage their learning skills by employing these interactive visualizations and may perform trickier tasks in the near future. Soon we would be witnessing our personal robots interacting with humans online and asking for help and acquiring new skill set with the whole world, how cool is that o-O
Here is the project video from Human-Centered Robotics Lab
Source: University of Washington