01
Human interaction through foundation models
Demo development focused on using foundation models to interpret what a person asks and adapt the robot’s task in response.
Technology
GalacNo is developing software to connect existing site systems with robots, sensors and new equipment. The intended workflow starts with a request and ends with evidence a person can review.
Demo development
We have been building demos of human interaction through foundation models, robot mapping and navigation in environments shared with people. Alongside these, our work includes a system to coordinate robots and humans, a digital twin of the activity, and an orchestration layer within that twin.
01
Demo development focused on using foundation models to interpret what a person asks and adapt the robot’s task in response.
02
A robot mapping its surroundings to build a spatial view of the environment.
03
A robot navigating environments shared with people, responding to movement and changes around it.
04
A system coordinating work between robots and people in the same environment.
05
A digital representation of what is happening in the physical environment, bringing robot and human activity into a shared view.
06
An orchestration layer within the digital twin, connecting that shared view to the coordination of robots, people and tasks.
The workflow
These functions describe the product being developed. Integration and behaviour must be validated for each site and job.
Bring together the work request, asset records, maps and available sensor data. Identify missing information before planning.
Propose the route, equipment and measurements, within the site’s access rules and operating limits.
Ask the authorised site operator to review the scope, equipment and limits before work starts. Changes outside that scope need a new decision.
Connect the approved job to the selected robots, sensors and site systems. Pause and hand back to an operator when the job cannot continue within its limits.
Link captured data, timestamps, equipment status, approvals and exceptions to the original request for operator review.
People retain control
Before work
The site nominates the person authorised to approve the job. They review its scope, route, equipment, operating limits and stop conditions before execution.
During work
The intended handback conditions include an obstacle, equipment fault, lost connection, insufficient sensor data or a request outside the approved scope. The configured stop or recovery procedure is agreed with the site; the operator decides whether to revise, resume or end the job.
After work
The intended record includes the original request, approved plan, approver and time, equipment and software versions, captured data, interventions and completion status. The operator accepts the result or requests more work.
Technical detail
“An OS for the physical world” is a supporting analogy: a common software layer connecting systems and coordinating work. The product is robotics software, and the scope of each integration matters.
Sensor inputs and asset records contribute to a shared site model. The research architecture calls these components Hive-Mind and Sentient Node. Data freshness and missing coverage need to be visible to the planner and operator.
The orchestration layer, previously described as Fulcrum Fabric, connects a job plan to equipment adapters. Commands need to respect the approved scope, device capabilities and operating limits. Manipulation is a separate capability to validate where a job requires it.
The interface presents requests, proposed plans, approvals and exceptions. The earlier Digital Teammate and Omni-Sense concepts inform interaction and monitoring; they do not establish a safety certification or replace site procedures.
Simulation can help evaluate a plan before a controlled test. Recorded events should support review and replay. Software changes need version records, validation and an agreed recovery process before use at a site.