Skilligent Robot Learning System
Skilligent has developed a trainable control system for multi-task autonomous robots. The unique feature of the product is that it enables the robots to learn procedures and skills directly from their human users while maintaining a social contact with them. Prior to Skilligent, the robots had to be programmed by engineers for each specific application. The high costs of the robot programming and the complexity of the systems integration scared off most potential non-professional robot users. The introduction of Skilligent software provides the users with an easy way to train their robots to perform a wide range of assignments without the need to program or configure the robots. The Skilligent control system now enables robots to be trained in the same way as home pets.
The software analyzes the user's gestures, looks at the objects presented by the user, listens to the sounds and tries to guess what the robot is supposed to do. Through trial and error, the robot understands what needs to be done and associates learned behaviors with stimuli used by the human. Over a few training sessions, the robot refines its understanding of the domain, the procedures and skills required to serve the user.
Videos demonstrating the product's capabilities
This series of videos demonstrates the basics of the robot training process. The videos show a robot being trained to execute procedures of increasing complexity.
The fist goal is to train a robot to go to a certain place in a room, pick up an object there and then drop the object into a bin located somewhere else. After that, the robot is trained to perform a "sorting" task - dropping different objects into different bins.
A sample mobile robot shown on the videos is equipped with a 6 DOF robotic arm manipulator.
Video 1: A robot performs a learned procedure
The first video demonstrates how a pre-trained robot autonomously performs a learned procedure.
The robot has been pre-trained to do the following:
The robot can reliably perform the procedure over and over again. Skilligent software uses a built-in robot vision system in order to locate objects, look for landmarks, navigate around the place, estimate object's position and control the robotic arm manipulator.
If a landmark, an object or a bin has been moved from their original places, the robot is still capable of successfully completing the procedure.
Video 2: A robot training session
The next video shows how the robot was trained to perform the procedure shown in the previous video.
During a training session, the robot needs to learn how to navigate to a certain place, pick up an object there and drop the object into a bin located somewhere else.
The training does not require any special technical skills from the trainer. The user doesn't touch a keyboard or joystick throughout the training.
Video 3: Learning a condition
Let's train the robot execute a procedure which is more complicated than the previous one. Imagine that the user wants the robot to learn how to drop different kinds of objects into different bins (a sorting task). The robot needs to learn a condition - how to behave differently depending on the observed situation.
Video 4: Learning a task hierarchy
Once a robot has learnt a procedure, it is capable of reusing the procedure as a step in a larger procedure.
Imagine that a user wants the robot to perform the following cycle:
As you see the new procedure uses a previously learned procedure as one of the steps.
Skilligent software can learn task hierarchies of nearly unlimited depth.
Videos 5 and 6: Learning a skill
A skill is a low-level control rule which can be used as a step in a procedure. Technically, a skill is set of control loops (similar to PID loops) working together in order to achieve and hold a certain target. Skilligent software is capable of learning skills from robots users.
Let's take a look at how a robot is being trained to use its robotic arm manipulator.
The first video shows how a pre-trained robot executes a skill.
The second video demonstrates how the skill was actually learnt by the robot:
© Skilligent Inc |