Robots and the Human

CSCI 8970 – Colloquium Series – Fall 2010 – Eight Event

Robots and the Human

Monday, November 1, 2010

While Robotics is not as advanced in the United States as it is in Japan, the University of Minnesota had the privilege of hosting Oussama Khatib, a professor and researcher at Stanford University whose job specializes in increasing the sensitivity of robots to increase their reactions to and safety in relation to the environment. Robots are complicated in both their hardware and software development. Dr. Khatib began his presentation by showing a video of a machine trying to feed a person over and over again and how without proper engineering the robot hindered rather than helped the human.  Currently robots such as the Puma 560 have a payload 22N or 1.16 which is much higher than the human average payload of 62n or 0.04. Dr. Khatib is currently working on robots that have a safety payload even better than a human at 60n (DM2) and 33n for the S2p robot.

To achieve this, Dr. Khatib believes it is important to improve a robot’s perception. Robots are increasingly being aware of where things are in the environment but they still cannot interact with the environment. Can we really go beyond walking and deal with the interactivity with the environment? To do so, we need for whole body control. This is the challenge faced by Dr. Khatib’s team. Part of the challenge relates to sensing and perception of the real-time, unstructured world, and planning, control, and skills by giving many degrees of freedom to the robot.

To improve the robots interaction with the environment, Dr. Khatib has focused on developing a human-friendly robot design. Usually it is debated that the robot can either be safe or perform adequately, yet why not improve both? Dr. Khatib attempted this by developing the DM2 Distributed Macro-Mini Actuation, a robot with two actuators. The later developed the S2P or Stanford Human-Safe Robot. Trying to build an arm with the constraints that face a human, they were able to achieve high control and dynamic. Their robot has an arm of 1.2 kilograms, instead of the 40 kilograms of the PUMA which cannot be touched. Now the robot has shoulders, they also added muscles, but muscles are weak, without the aid of a motor for strength. They are also developing a skin to reduce collision damage as well as including proximity sensors.

To illustrate mobile manipulation and human guided Motion, human robot interaction, Dr. Khatib showed a video of two humans interacting. Romeo and Juliet (1993) – had a video of the robot ironing a shirt. He stressed the need for the robot to be compliant and move with the human. In 1997, Honda released the P2 and Dr. Khatib was challenged with building the capabilities given to Romeo and Juliet into Asimo. It was exciting but a very difficult problem. In the end, the achieved, humanoid robot control. When you touched the environment you bring a lot of new constraints. They worked on creating a robot that can develop the skills. Yet, developing inverse schematics is incredible difficult for robots. They tried to solve this problem through goal directed control exhibited by humans. In order to create the behaviors for the robot – they go to the human and model him. Acquired models of the muscles and see the criteria humans used.  As humans learn a skill, a task, humans find ways to make more efficient the use of our muscles. Humans discover the physio-mechanical advantage. Humans adjust the body so that all muscles do not saturate.

The solution for him laid in E = cm^2. Like humans who lifting their arm only 42 degrees when drinking a cup of coffee because that is the minimum energy needed, they used this ideas to create a program that he changes the place a ball is and the robot follows – doing all the movements gently. The program copied the motion characteristics, not the trajectory. They applied this fluid motion to Asimo. Also, by building this mapping tool they develop a useful tool also for humans. Humans when using their bodies for an activity over time find the largest acceleration lines. That’s why we do not throw objects in a straight line. Attempting to create a human motion atlas they went to South Africa and use a motion capture and motion reconstruction software on two dancers (one male and one female). The use of this program can help improve training.

Some constraints they have worked to solve include joint limits or self-collision, the need for internal and external constraints and the need to find distance between objects really quickly in real time. To illustrate some of these problems, Dr. Khatib displayed a Chai 3D simulation of skin and a hand. From haptics they can tell the robot to perform. The robot, through the software has been able to learn skills. They are currently working in improving the robots reactions to incoming objects, for the robot to be able to feel an obstacle coming and reacting to the object by moving away.

In terms of integration of locomotion, usually contact made the robots lose their balance. Humans need to find balance all the time other way they fall. With this information at hand, they were developing a model for each part and then tried to integrate it. Now they are using a framework were all the body is taken into account at the same time. The problem is that most robots do not have a tore control system Asimo was not tore control which increased the difficulty – robots are moving to tore control. Dr. Khatib showed a diagram of the robot’s architecture and spoke about the importance for robots to have a plan so that they do not have to make a whole analysis every time. Robot needs a plan that it can adapt. In the end of the presentation, Dr. Khatib showed videos of if group teaching Asimo how to grab things and walk. The presentation was incredibly exciting. The room was full and one can only wonder where robotics will take us next. Hopefully the United States will increase its research and study of the field.