Open-RMF execution

eplansys_rmf_bridge runs an epistemic policy on a robot fleet managed by Open-RMF. It is an ActionExecutorClient: it accepts a dispatched action, submits the corresponding RMF task, waits for completion, and calls finish() with the outcome the robot observed. plansys2_msgs/ActionExecution carries an outcome field for this, and the policy branches on it.

The bridge is maintained in its own repository, eplansys-rmf. Open-RMF is a substantial dependency, and requiring it from eplansys would render the planning stack unusable for installations that do not operate a fleet. This is the arrangement plansys2_aletheia_plan_solver already uses for its external binary.

EPDDL is grounded by plank and planned into a branching policy; the executor dispatches an action to eplansys_rmf_bridge, which submits an RMF task; the robot executes it and the observation returns as an outcome that updates the epistemic state.

Fig. 9 Where the bridge sits between the policy and the fleet

The loop closes at the bridge: it is the only component that both dispatches into the fleet and receives what the fleet observed. Everything above it is the planner’s, everything below it is Open-RMF’s, and the outcome edge is the one that makes the execution epistemic rather than a remote control.

Division of responsibility

The two systems decide disjoint things. Open-RMF holds no representation of knowledge, and ePlanSys holds no representation of shared physical resources.

System

Determines

ePlanSys

what has to be found out, which agent should act, and which agents may come to know the result

Open-RMF

the allocation of floor space, lifts and doors, and which robot is available to perform the work

What a performer does

The process reads its task map and turns every action name in it into one performer. What that performer does when the executor dispatches it depends on the entry:

  • An action that moves a robot submits an RMF task pinned to the robot its agent is bound to, waits for the fleet to report it done, and finishes.

  • An action marked local is a speech act. RMF has no representation of saying things and no robot moves to say them, so nothing is submitted and the performer completes after its stated duration. The epistemic effect is entirely the planner’s.

  • An action marked sensing finishes with an outcome token. This is the argument the policy branches on, and the only reason the bridge is more than a remote control.

Reference scenario

The survey domain: three robots and a site that may be contaminated, under a goal of three conjuncts. The scout is required to find out whether the site is contaminated, the relay to come to know the result, and an observer not to. The planner declines to broadcast and uses a private channel instead, since broadcasting would falsify the third conjunct. Executing it over an RMF fleet, with RMF driving the robots, is the intended demonstration of the bridge, and is described in Running the survey demo.