RESILEX

RESILEX research note

HEX and excavation

The first autonomous excavation loop

A narrow RESILEX scenario connecting a digital task, human approval, machine execution, terrain measurement and result acceptance.

Excavation autonomy is not complete when a bucket trajectory ends. The resulting terrain must be measured and compared with the intended physical outcome.

01

Why start with earthworks

Earthworks combine heavy physical action with a result that can be measured. The terrain changes continuously, the machine must understand its position and work zone, and small errors can affect every operation that follows.

This makes excavation a useful first problem. It is narrow enough to model, but broad enough to expose the connection between planning, machine control, perception and acceptance.

  • Changing terrain
  • Heavy tool
  • Measurable geometry
02

The six parts of the loop

The loop begins with a digital task that defines the target surface, limits and acceptance criteria. A person reviews the task. HEX then receives a machine-level operation and performs the work inside the approved zone.

After execution, the terrain is measured. The system compares the actual surface with the target and either accepts the work, requests investigation or prepares another pass.

  • Task
  • Approval
  • Execution
  • Measurement
  • Verification
  • Decision
03

What HEX has to provide

HEX needs dependable localisation, a clear machine interface, control of the excavation tool and a way to report its own state. It also needs safe manual and remote modes so development does not depend on full autonomy from the first prototype.

Tracked and six-legged arrangements remain open design studies. The choice should follow stability, mobility, mechanical complexity, serviceability and simulation results.

04

What counts as evidence

A convincing result is not a rendered animation of a machine moving. Evidence should show the task definition, the executed motion, the measured terrain, the error relative to the target and the decision produced from that error.

The first digital model and simulation should therefore preserve these links. A later physical prototype can then test the same structure in the real world.

  • Reproducible task
  • Measured error
  • Clear acceptance decision