Articles, Blog

AI Planning for Robots

December 1, 2019

AI Planners are used in the generation of
action sequences in robots such as the Freddy 2 Hand/Eye Assembly Robot displayed
here in the National Museum of Scotland in Edinburgh. This version of Freddy from
four decades ago had a large, fixed mounted arm with sophisticated touch
sensitive grippers, and a side mounted camera to look at its world. Instead of
the robot moving to reach objects, the platform underneath the robot was shifted
instead. Freddy was demonstrated on flexible assembly tasks involving simple
toy cars and boats. The video playing is taken from a sixteen millimeter film of a
Freddy project demonstration made in 1973. Freddy had to identify jumbled up parts
via vision, unscramble the parts, and then assemble them to create the required
product. Some parts in the area didn’t belong to the target assembly. In one
project, the Edinburgh Nonlin AI Planner was used to generate object construction
and arm movement sequences for Freddy. In its time, Freddy was one of the most
sophisticated robots in the world. The Stanford Research Institute Problem
Solver – STRIPS – is one of the best known and most influential AI Planners, and
you’ll learn about it on this course. Despite it being created over 40 years
ago, it was already in use to control the activity of a robot called Shakey at the
Stanford Research Institute, which is now called SRI International. STRIPS had many
interesting features including using theorem proving techniques to reason about
the state at points in the plan. STRIPS gave us a representation of actions
with preconditions and effects. It had a mechanism called MACROPS, for
generalization of its operators. And it included execution support facilities
to deal with partial failures. Shakey is now on display at the Computer
History Museum in Mountain View, California. As we saw, Freddy included an early robot arm,
but as you can see here, technology has come a long way. For example, this modern
research robotic arm has many more capabilities. Obviously, there’s a lot
more to robotics than just activity planning and there are significant
challenges in terms of vision, mechanical issues, and spatial reasoning including
locating the robot, and tracking other objects. But planning is a key aspect of
intelligent behavior. Not every robot has the kind of activity planner that we are
describing on this course. That is, one which can plan from first principles. But
robots typically do have plans, perhaps precompiled ones, that they use to control
their behavior. AI Planners have been used in much more recent robots also, of
course, such as in some humanoid robots and the robots used in the Annual RoboCup
Soccer League Teams. Here, planning may have to cover the coordination of
different robots with different roles, and the robots themselves can have planners on
board that adjust their behaviors dynamically. But planners have been used
in different kind of robots too. For example, an on board planner and reactive
execution agent was used to autonomously control the NASA Deep Space One spacecraft
for periods on its flight to rendevous with a comet. AI Planners have been used to provide
adaptability and flexibility in robots in the past and in the present, in the
factory, in the home, and in outer space. And its an area in which they will be put
to good use tomorrow. We hope you’ll be part of creating this future.

You Might Also Like

1 Comment

  • Reply Vinicius Pires January 16, 2015 at 1:33 am

    Great video 🙂

  • Leave a Reply