Changed response
The same command produces a different movement.
Explore adaptation to actuator calibration, altered wheel response or a changed payload, using the observations available to the robot.
Pragma Research / Robotics
Give your robot a brain that learns from experience.
Cadence is a deep real-time brain that learns from experience. We are seeking robot makers and integrators to put it to work on a concrete adaptation or recovery task.
The ambition
A brain that improves through a machine's working life.
A robot's body and surroundings change. We want its experience of those changes to improve its behavior.
Cadence's working software core demonstrates learning in simulation. Our proposed physical programme starts with one body and one task where adaptation has measurable operational value.
The product is the Cadence brain and runtime. A focused integration gives your team a practical way to evaluate that larger architecture.
Find the right first task
Where should experience change the outcome?
Bring a recurring problem your team can reproduce. Together, we can test whether learning from operation reduces the work needed to recover.
Recurring recovery
One failure keeps calling an operator back.
Choose a bounded recovery behavior and measure whether experience improves completion or reduces human intervention.
Skill retention
Learning another behavior should preserve useful skills.
Evaluate each learned task from the same brain state. Measure retained performance and the resources used to achieve it.
These are proposed evaluation areas. Task selection starts with the failure mechanism and your team's operational priorities.
Proposed demonstration
A wheel weakens. The robot learns how to move with it.
A small rover makes the adaptation question visible.
The proposed experiment starts with a robot reaching targets. One wheel's response changes. Cadence receives ordinary motion feedback and learns from the consequences of its commands.
The test compares recovery with frozen control and competent learning baselines, then measures performance when the original condition returns. A successful simulation would support a focused physical evaluation.
This is a proposed experiment. Existing measured learning evidence is available from the Doom Basic simulation.
Inspect the existing evidence →A paid evaluation
One task. A shared test. A deployment decision.
01
Define the task
Your team identifies the operational problem and technical sponsor. Together, we agree the observations, available experience and success criteria.
02
Connect the brain
Integrate Cadence with a simulator or robot interface. Specify which decisions it controls and measure the full sensing-to-command path.
03
Measure useful work
Compare against your existing stack and appropriate adaptive baselines. Count successful tasks, interventions and recovery experience alongside learning and compute costs.
04
Choose deployment
Review the integration and reproducible evidence together. Useful results can lead to a supported runtime and embodiment license.
Inside Cadence
A brain built from systems that observe themselves.
Cadence instantiates observer-like, self-reading software patches with local state, ports, records and feedback. Settlement repairs their activity; experience can change their retained relations.
Choose a layout for the task: flat patches read sensors directly, ordinary state-coupled populations form intermediate representations, and recursive observers also read exact prediction errors. All three use the same repair and qualification procedure. Measure useful behavior alongside the complete control-loop cost.
Work with Pragma Research
Bring a body. Bring a problem worth learning.
We welcome robot manufacturers, integrators and research teams with a clear task and an accessible control interface.
For an initial conversation, tell us the robot, the recurring problem and how you measure success. We can demonstrate the software, assess the fit and scope a paid evaluation with your technical team.
The proposed commercial path is evaluation, production integration and recurring licensing of the runtime and embodiment package.