Integrated Task and Motion Planning in Robotics - Robotic Sea Bass

In the previous post, we introduced task planning in robotics. This field broadly involves a set of planning techniques that are domain-independent: That is, we can design a domain which describes the world at some (typically high) level of abstraction, using some modeling language like PDDL. However, the planning algorithms themselves can be applied to any domain that can be modeled in that language, and critically, to solve any valid problem specification within that domain. The key takeaway of the last post was that task planning is ultimately search. These search problems are often challenging and grow exponentially with the size of the problem, so it is no surprise that task planning is often symbolic: There are relatively few possible actions to choose from, with a relatively small set of finite parameters. Otherwise, search is prohibitively expensive even in the face of clever algorithms and heuristics. Bridging the gap between this abstract planning space and the real world, wh

In the previous post, we introduced task planning in robotics. This field broadly involves a set of planning techniques that are domain-independent: That is, we can design a domain which describes the world at some (typically high) level of abstraction, using some modeling language like PDDL. However, the planning algorithms themselves can be applied to any domain that can be modeled in that language, and critically, to solve any valid problem specification within that domain. The key takeaway of the last post was that task planning is ultimately search. These search problems are often challen