Curiosity Saved The Robot
Due to several cyber attacks, RobotRabbi was down intermittently a few weeks. We are now back, more stable and safer than ever. We are excited to share with you a new design for RobotRabbi. You will notice many new features to share and sign up for new posting. As we begin anew, it makes me think about robots in their own infancy.
At a recent TEDx Talk in Cannes France, Pierre-Yves Oudeyer presented a novel concept of using robots to better understand the ways newborn mechanisms learn (see above). As any parent knows, a child’s brains and hands are used to fabricate and model the things a baby wants to better understand. Play is critical to development during a time when the brain has the greatest plasticity. Oudeyer explains that scientists also use fabrication to build new knowledge of the world around us. Scientists build large scale aquariums to understand ocean behavior and construct large computer simulations to understand spiral galaxies.
The idea of fabricating baby robots and providing them with the tools to learn is the central idea of this talk. Robots are given the tools to create their own experiments and exhibit forms of cognitive development. Curiosity is a large focus of robot learning. Pierre-Yves explains that children use curiosity as a learning mechanism but they do it in a very structured way. His team built robots that could learn, discover, and set their own goals.
The first experiment shown has two robots with quadraped bodies in a playground environment. Actions are performed, effects are logged into the internal database, and the robot tries to detect irregularities in the effects. This gives the robots the ability to make predictions about future states. Robot brains choose experiments that they think will provide the most progress in their internal predictive algorithms. This allows the bots to gain new skills but also brings in a self-organization that occurs between the robot and its environment. Eventually each robot creates a system of vocal interaction with the other robots. In the video it sounds like a cross between puppy and kitten mewling, and the sounds are transmitted between bots until they’re all making the same sounds.
Oudeyer emphasizes the link between the robots learning, curiosity, the robot body and the environment. He says that changing the body but keeping the same learning mechanism will create different cognitive learning stages, in a different order. The entire talk is fascinating and covers not just learning but also communication and languages. At the end of the talk Oudeyer introduces Poppy, the open source 3d printed humanoid robot (similar to Intel’s Jimmy, see my previous post). He says that this robot will allow every lab, school or fabrication location to join the scientific exploration of robotic learning. Poppy is worth a few articles in and of itself, just on the immense scale of the project and the effort and skill required to build and program one of the bots.
As we look to robot learning for a paradigms of our own human development, the converse is also true. Robots will eventually outgrow the lab and will be self-learning modules specific to their tasks gaining expertise with experience.