Learn the procedure
A human demonstration can be converted into an explicit sequence of sub-tasks, constraints, and reusable atomic actions.
Procedural Runtime Intelligence
Teach Once.
Execute Immediately.
Adapt at Runtime.
Adora Robotics is building Procedural Runtime Intelligence (PRI) for general-purpose robots. Instead of training every task into a new policy, PRI keeps the task explicit: a structured procedure the robot can execute, inspect, modify, and reuse.
Why PRI
Robots should learn tasks as procedures, not memorize them in weights.
A task is more than a trajectory. It contains steps, goals, constraints, choices, and recovery logic. PRI represents that structure explicitly, so the robot can reason over the task itself instead of treating behavior as an opaque output of a trained policy.
Execute with perception
The procedure stays stable while perception and control continuously update object poses, motion, timing, and local execution parameters.
Change the task at runtime
Because task logic is explicit, new user instructions or changing conditions can modify what the robot does without retraining the task.
How it works
From demonstration to execution.
1
Teach
A person demonstrates the task naturally.
2
Understand
PRI extracts the procedure, sub-tasks, constraints, and reusable actions.
3
Execute & Adapt
The robot executes the procedure and adjusts to runtime changes.
No task-specific policy training.
No retraining for every new task variation.
A different abstraction
A task is not a frozen trajectory.
PRI separates the procedure from the moment-to-moment execution details.
Interactive
The user can change the requested outcome or task details, including after execution has begun.
Explicit
Task structure can be inspected, edited, debugged, and improved step by step.
Reusable
Procedures can reuse existing atomic actions and procedural fragments instead of learning every task from scratch.
Demonstrated tasks
From simple manipulation to multi-step work.
Our prototypes have executed a growing range of real-world manipulation tasks using the same procedural approach.
Object sorting & placement
Pouring coffee beans
Grocery shelf stocking
Cake decoration
Hamburger assembly
Coffee preparation
Milk tea preparation
Multi-step task adaptation
New English-language demonstrations are in production and will be added here.
Beyond planner + skill
More than choosing among predefined skills.
Traditional planner-and-skill systems assume engineers have already defined the task capabilities. PRI is designed to acquire new procedures from demonstration, keep them explicit at runtime, and compose them from reusable actions. The goal is not only to select known skills, but to let robots acquire, modify, and reuse new procedures without training a new task policy.
Our goal
A general-purpose robot should not need a new model for every new job.
We are building PRI as a runtime intelligence layer that lets ordinary people teach robots what to do — and lets robots turn those demonstrations into reusable procedural intelligence.