VISION + POSITION
Find the next contact.
Estimate the instrument’s geometry, string location and the hand’s relation to the target. Changing viewpoint should not erase what has been learned.
FLAGSHIP RESEARCH CHALLENGE
Learning to play.
Learning to learn.
One instrument. An exacting test of perception, dexterity, prediction and experience.
THE AMBITION
A robot that earns
its next note.
The objective is for a robot to acquire guitar-playing skill through its own bounded practice and sensory feedback.
That means learning how contact, position and movement affect sound; using the difference between an intended note and the measured result to improve the next attempt; and retaining useful skill across practice sessions.
Autonomous practice still requires a defined task, an instrument, body calibration and independently enforced operating limits. Prior developmental experience must be disclosed. “Learning by itself” describes the intended learning process, not an absence of engineering, training or supervision.
Research objective. The current Life Core work is a software research prototype; no autonomous guitar-playing robot has been demonstrated.
Sound makes the
outcome observable.
The instrument creates a demanding connection between physical action and an independently measurable result.
TOUCH + FORCE
Make a clean note.
Explore the relationship between finger placement, pressure and the resulting sound, while respecting the physical limits of the hand and instrument.
TIMING + COORDINATION
Bring two hands together.
Coordinate fretting and plucking across a sequence. The next action depends on both current contact and the phrase that follows.
HEARING + MEMORY
Learn from the result.
Use measured pitch, onset and unwanted noise to evaluate an attempt. Test whether later practice improves and earlier skills survive.
WHY THIS BENCHMARK
A familiar challenge.
A serious control problem.
Musical performance already has a place in robotics research. RoboPianist studied simulated dexterous hands across a repertoire of piano pieces. Subsequent work transferred learned piano policies to physical hardware. These studies support music as a manipulation benchmark; they do not establish a general-purpose musician. [1, 2]
Guitar adds a different configuration of continuous string contact, asymmetric hand roles and audible contact errors. Our thesis is that these properties can expose the quality of a learning loop more clearly than a rehearsed demonstration alone. That is a proposed benchmark rationale, not a claim that guitar is universally harder than piano.
A successful phrase is only part of the evaluation. The stronger test is the learning curve: measured practice, error recovery, retention and performance on unfamiliar phrases or changed conditions.
The path to a
credible demonstration.
Proposed capability gates, ordered by dependency. These are research milestones, not completed achievements or delivery dates.
Stable perception
Identify relevant instrument features and estimate contact geometry on unseen views.
One action, one consequence
Predict and measure how a controlled movement changes contact and sound.
A reproducible clean note
Learn within fixed limits; compare progress against scripted and conventional learning controls.
A learned phrase
Coordinate hands and timing, measure recovery, and retain skill across sessions.
Transfer beyond rehearsal
Test unfamiliar phrases, changed conditions and a separate physical task before claiming broader transfer.
WHAT WE WOULD MEASURE
Count the practice.
Measure the progress.
Evaluation should record real interactions, prior demonstrations, simulation work and human interventions. Musical outcomes should include note correctness, timing error, unintended string contact and the ability to recover after a controlled disturbance.
Retention and transfer need separate tests, held conditions and an unchanged scoring rule. Compute, memory and latency belong alongside capability. No individual metric is a substitute for a complete physical result.
How this connects to the platform thesis ↗BUILDING THE NEXT CHAPTER