Track D · research agenda

Physical

How do Route, Ground and Assure apply where errors have physical consequences?

The same Route, Ground, Assure stack where errors have physical consequences. Every claim will name its testbed stage and report sim-to-real gaps.

Directions

  1. D1: swarm coordination under degraded conditions.
  2. D2: assurance for learned autonomy (runtime monitors and safety shields).
  3. D3: language and agents as robot interfaces, with verification gates.
  4. D4: assistive robotics with demonstrable safety.

Picture

  1. 1. Simulation: ROS 2, Gazebo (Planned)
  2. 2. Hardware-in-the-loop: Flight controller or board with simulated environment (Planned)
  3. 3. Controlled indoor trials: Small, safe, repeatable (Planned)
  4. 4. Outdoor field trials: Only with permissions and a safety pilot (Planned)
View as table
Testbed ladder
StepToolsStatus
SimulationROS 2, GazeboPlanned
Hardware-in-the-loopFlight controller or board with simulated environmentPlanned
Controlled indoor trialsSmall, safe, repeatablePlanned
Outdoor field trialsOnly with permissions and a safety pilotPlanned

Digital versus physical

LayerDigitalPhysical
RoutePolicy-driven model and provider selectionPolicy-driven planner, controller and fallback selection
GroundProvenance-first knowledge graphWorld model with sourced observations and versioned maps
AssureQuality contracts before an answer shipsRuntime monitors and safety shields before an action executes

Planned metrics

Values stay Pending until we publish logs.

Mission completion under injected faults

Share of missions completed after planned faults.

Value: Pending

Time to re-plan after losing an agent

Seconds from agent loss to a new valid plan.

Value: Pending

Unsafe-plan interception rate

Share of unsafe plans stopped by the shield.

Value: Pending

False-block rate

Share of safe plans incorrectly blocked.

Value: Pending

Connects to

Route · Ground · Assure · Foundation