System

Software stack

One system, two competitions. The modules below are shared between the Mart Humanoid Robot Challenge and RoboCup@Home; only the task layer differs.

Both competitions we enter — the Mart Humanoid Robot Challenge and RoboCup@Home — are prepared as a single service-robot system rather than two separate efforts. The modules below are shared; only the task layer differs between the two.

Modules

Bringup and system integration

Bringing the platform up reproducibly: drivers, transforms, time synchronisation and the launch layer that lets any member start the whole stack from a clean machine.

Localization and navigation

Mapping the arena, localising in it and moving through a furnished room without contact — the base capability every task in both competitions is built on.

Perception of people and objects

Detecting and tracking people, recognising the objects a task refers to, and estimating the poses needed to grasp them.

Manipulation and pick-and-place

Grasp selection and arm motion for the object set of each task, including handovers to and from a person.

Task planning and state management

Turning a spoken or written instruction into an ordered plan, and keeping the robot's belief about the task consistent while it executes.

Failure detection and recovery

Noticing that a step did not do what it was meant to, and recovering without human intervention — the difference between a demo and a run that scores.

Sim-to-real

Training and evaluating policies in simulation, then transferring them to the physical platform with the gap measured rather than assumed.

Repeatable evaluation

A fixed mock arena, controlled lighting and a scripted run protocol, so that a number measured this week can be compared with one measured last month.