DEVELOPMENTAL INTELLIGENCE FOR ROBOTS
Intelligence,
earned through
experience.
Building the learning foundation for robots that observe, adapt and carry their experience into the next challenge.
01 / OUR CONVICTION
The next skill should start with the last experience.
A robot’s useful knowledge should grow as it encounters the world.
Our research examines how perception, prediction and memory can work together so that what a robot experiences changes how it approaches the next unfamiliar situation. We call this developmental intelligence.
Why we’re building it ↗02 / THE RESEARCH FOUNDATION
See. Anticipate.
Retain. Adapt.
Life Core is our research programme for Salience-Gated Predictive Memory: a foundation for experience that remains useful beyond a single encounter.
Perception
Gather evidence about the world and revisit uncertain observations.
Prediction
Use acquired experience to form expectations that can be tested.
Memory
Retain useful observations and preserve acquired state across sessions.
Adaptation
Measure whether new experience improves the next decision.
Our long-term challenge is a robot that learns to play guitar through its own practice.
Vision locates a string. Touch regulates contact. Movement shapes the note. Hearing closes the feedback loop. Progress requires those capabilities to develop together.
For us, guitar is a rigorous test of learning: whether a system can acquire a demanding physical skill, recover from mistakes and carry what it learns into unfamiliar variations.
Flagship research objective. Autonomous guitar playing has not yet been demonstrated.
04 / RESEARCH WITH ACCOUNTABILITY
Progress that can be examined.
We distinguish a working mechanism from a useful capability.
Our research record includes bounded memory, repeatable continuation and a measured prediction-learning result. It also records where recognition has not improved and which physical capabilities remain to be built.
Read the research brief ↗BUILDING THE NEXT CHAPTER