PHASE SHIFT
How close isrobotics to itsChatGPT moment?
From impressive demonstrations to useful intelligence in the physical world.
Provisional editorial estimate
An editorial index, not a probability.
The current assessment
MethodologyPHASE’s initial AI-assisted estimate is 40%. The strongest evidence concerns translating instructions into actions and extending learned behaviours across tasks. Evidence reviewed here does not establish affordable, broadly accessible robots that work reliably for long periods in unfamiliar environments. The estimate deliberately discounts demonstrations without independently established intervention rates or deployment economics. It is not a benchmark score, a prediction, or a probability.
The percentage is a subjective editorial index. It does not mean this fraction of the engineering work is complete, or that a breakthrough has this probability. Component scores are editorial judgments too.
Generalisation40/100
Transfer across tasks and embodiments is progressing, but the breadth of genuinely unfamiliar tasks remains uncertain.
Weight: 20% · Source 1 Source 2Manipulation50/100
Reported whole-body and dexterous results are meaningful, while difficult multi-finger tasks remain uneven.
Weight: 15% · Source 1Language → Action65/100
High-level instructions can drive sequences of physical actions in reported systems; this is the strongest component in our judgment.
Weight: 15% · Source 1 Source 2Unseen Environments35/100
New-home experiments support transfer beyond the training location. They do not establish reliable performance in arbitrary homes.
Weight: 15% · Source 2Long-Horizon Autonomy25/100
Reported multi-step sequences and recovery are encouraging. Several-minute performance is not evidence of unattended day-long operation.
Weight: 15% · Source 1Reliability20/100
Published task outcomes remain uneven. The reviewed evidence does not establish the repeated low-intervention reliability required by our endpoint.
Weight: 15% · Source 1Cost / Deployment35/100
On-device models and adaptation can reduce operational friction, but the reviewed material does not establish broadly accessible total deployment costs.
Weight: 5% · Source 1Evidence reviewed
- Gemini Robotics 2 brings whole body intelligence to robots
Google DeepMind · Published 2026-07-30 · Reviewed 2026-09-14
Primary developer announcement; reported evaluations and demonstrations, not independent verification. - π0.5: a Vision-Language-Action Model with Open-World Generalization
Physical Intelligence / arXiv · Published 2025-04-22 · Reviewed 2026-09-14
Primary research paper; older foundational evidence, not a new September 2026 event.
Reading history
Reassessments create new entries. Earlier judgments remain visible.
Initial estimate 40%
Provisional editorial estimate
Reasoning, evidence and methodology
PHASE’s initial AI-assisted estimate is 40%. The strongest evidence concerns translating instructions into actions and extending learned behaviours across tasks. Evidence reviewed here does not establish affordable, broadly accessible robots that work reliably for long periods in unfamiliar environments. The estimate deliberately discounts demonstrations without independently established intervention rates or deployment economics. It is not a benchmark score, a prediction, or a probability.
Generalisation · 40/100
Transfer across tasks and embodiments is progressing, but the breadth of genuinely unfamiliar tasks remains uncertain.
Manipulation · 50/100
Reported whole-body and dexterous results are meaningful, while difficult multi-finger tasks remain uneven.
Language → Action · 65/100
High-level instructions can drive sequences of physical actions in reported systems; this is the strongest component in our judgment.
Unseen Environments · 35/100
New-home experiments support transfer beyond the training location. They do not establish reliable performance in arbitrary homes.
Long-Horizon Autonomy · 25/100
Reported multi-step sequences and recovery are encouraging. Several-minute performance is not evidence of unattended day-long operation.
Reliability · 20/100
Published task outcomes remain uneven. The reviewed evidence does not establish the repeated low-intervention reliability required by our endpoint.
Cost / Deployment · 35/100
On-device models and adaptation can reduce operational friction, but the reviewed material does not establish broadly accessible total deployment costs.
PHASE SHIFT provisional v1
An editorial assessment of readiness for broadly useful, instruction-led robotics in unfamiliar environments. Zero means no general-purpose capability; 25 means mostly narrow demonstrations; 50 means varied tasks demonstrated with substantial operational limitations; 75 means repeated useful deployment across unfamiliar settings with limited intervention; 100 means broad, accessible and reliably useful general-purpose robotics. These anchors are judgments, not measured stages of a fixed engineering programme.
Assess seven dimensions from linked evidence, giving greater weight to transfer and reliability than publicity. Manufacturer demonstrations are attributed claims, not independent validation. The weighted component estimate is rounded to the nearest five points. Editors may override it with a documented reason. Reassessments preserve their sources, component judgments and methodology version. AI assists research and drafting; a human editor remains responsible. No automatic updates or forecasts are in operation.
- Gemini Robotics 2 brings whole body intelligence to robots
Google DeepMind · Published 2026-07-30 · Reviewed 2026-09-14
Primary developer announcement; reported evaluations and demonstrations, not independent verification. - π0.5: a Vision-Language-Action Model with Open-World Generalization
Physical Intelligence / arXiv · Published 2025-04-22 · Reviewed 2026-09-14
Primary research paper; older foundational evidence, not a new September 2026 event.