Embedding a Scheduler in Execution for a Planetary Rover

Abstract

Scheduling often takes place in the context of execution. This reality drives several key design decisions: (1) when to invoke (re) scheduling, (2) what to do when the scheduler is running, and (3) how to use the schedule to execute scheduled activities. We define these design decisions theoretically in the context of the embedded scheduler and practically in the context of the design of an embedded scheduler for a planetary rover. We use the concept of a commit window to enable execution to use the previously generated schedule while (re) scheduling. We define the concepts of fixed cadence, event driven, and hybrid scheduling to control invocation of (re) scheduling. We define the concept of flexible execution to enable execution of the generated schedule to be adaptive within the response cycle of the scheduler. We present empirical results from both synthetic and planetary rover scheduling and execution model data that documents the effectiveness of these techniques at enabling the scheduler to take advantage of execution opportunities to complete activities earlier.

Optimizing Parameters for Uncertain Execution and Rescheduling Robustness

“…As variance increases, there will be situations where an activity finishes late and forces the execution system to handle the next back-to-back activity. The existing execution system is based on fixed start times and allows limited delays of activity start times (Chi et al 2018). Delays cannot always be handled and may result in an inconsistent schedule and unscheduled activities.…”

Temporal Brittleness Analysis of Task Networks for Planetary Rovers

“…We run the brittleness analysis using a set of nine M2020 sol type task networks. These networks are representative of what is currently being investigated and applied to develop an onboard scheduler for the M2020 rover (Chi et al 2018).…”

“…Planetary rover plans must be carefully designed and generated on the ground to allow successful execution of tasks while meeting all the required timing constraints and being safe at all times. Accurately determining some of those timing constraints a priori, specially activity duration, is quite challenging though due to the the natural unpredictability of the environment (Rabideau and Benowitz 2017; Chi et al 2018).…”

“…We derive task networks from M2020 sol types (Jet Propulsion Laboratory 2018), the current best available data on expected rover operations during a Martian sol. Based on (Chi et al 2018), we define a task network as a set of activities A = {a i T C, D ...a n T C, D }, where: T C is the set of temporal constraints referring to the nominal duration T C i,dur , earliest start time T C i,est , latest start time T C i,lst , and latest end time T C i,let ; and D is the set of the activity's dependency constraints in the form of a j → a k , meaning a j depends on a k , i.e. the end time of a k must be before the start time of a j .…”