Any-Start-Time Planning for SIPP

Devin Wild Thomas, Solomon Eyal Shimony, Wheeler Ruml, Erez Karpas, Shahaf S. Shperberg, Andrew Coles

Keywords: SIPP, anytime, arrival time, TDSP

TL;DR: Relaxing SIPP to plan for any-start-time requires searching over arrival time functions, rather than scalar g.

Abstract:

The problem of navigation among moving obstacles, sometimes referred to as SIPP, is a problem in which the applicability of an action, such as moving to a particular neighboring location, can change with the passage of time. This means that the optimal plan, and its duration, can change depending on when execution begins. In practice, the execution start time is often unknown until planning completes or another agent gives the go-ahead. However, most prior work on SIPP assumes a known start time and thus that the found plans are always feasible.
In this paper, we relax this assumption and directly address the setting of any-start-time planning. We present a general-purpose data structure that compactly encodes the optimal plan as a function of start time, as well as planning algorithms that utilize it. In experiments on SIPP, we find that this data structure tells us how long our paths remain valid, while its overhead is minimal. We also find that, in any-start-time SIPP, replanning is insufficient for difficult problems. In contrast, our any-start-time algorithms can be quickly queried for the optimal plan once the start time is known, albeit at the cost of substantial precomputation time.