SharpaSingapore2026
CraftNet
Full specifications
5 of 14 fields stated| Field | Value | Source and caveats |
|---|---|---|
| Release | ||
| Released | 7 January 2026 | [A] "Fine Manipulation 2026-01-07" A |
| Input and output | ||
| Model type | not stated | Note Sharpa's own term: "CraftNet is the combination of System 0 and System 1, forming a VTLA (Vision-Tactile-Language-Action) model that translates high-level, multimodal inputs into continuous, fine-grained actions executed by dexterous humanoid robots". A |
| Inputs | camera language robot state touch | Sharpa's model name spells out vision, tactile, language and action: "CraftNet is the combination of System 0 and System 1, forming a VTLA (Vision-Tactile-Language-Action) model that translates high-level, multimodal inputs into continuous, fine-grained actions executed by dexterous humanoid robots". Language comes from the layer above: "System 2 supplies semantic intent and/or explicit language instructions to System 1 within CraftNet". System 0 receives it "along with tactile and proprioceptive signals (e.g. force and torque)". A |
| Controls | arms and hands | "System 1 generates coarse motions, including upper-body movements and pre-contact hand poses"; "System 0 produces human-level, fine-grained actions in the last millimeters of contact". A |
| Degrees of freedom controlled | not stated | |
| Size and speed | ||
| Parameters | not stated | |
| Control rate | approximately 100Hz | The fastest of the three layers, System 0, the fine-motor layer: "It operates at ~100 Hz". Sharpa gives all three: "approximately 1 Hz for System 2, 10 Hz for System 1, and 100 Hz for System 0". A |
| Where it runs and what it is built on | ||
| On robot or cloud | not stated | |
| Base model | not stated | Note The reasoning layer uses an unnamed open-source model: "System 2 leverages an open-source VLM pre-trained on large-scale Internet data". A |
| Openness | ||
| Access | not stated | Note "Sharpa will share updates on CraftNet in phases". B |
| Licence terms | not stated | |
| Licence | not stated | |
| The developer’s own claims | ||
| Learns new tasks | needs retraining | Developer claim: "Most of the application-related training is performed on System 1 using public or private domain training data"; "System 1 and 0 are jointly trained and fine-tuned using high-quality teleoperation data in the post-training phase". A |
| Works across robots | not stated | Note Sharpa says CraftNet's actions are "continuous, fine-grained actions executed by dexterous humanoid robots". A Yes only if the developer itself says the model runs on different robot models, for example robots from two manufacturers or two of one manufacturer’s models. Several units of the same robot do not count. |
How to read the figures
- 50HzA
- The letter in the grey square after a figure points to the source the figure came from. Pages with several sources carry several letters.
Notes
- CraftNet has three systems: System 2 (the Reasoning Brain, a vision-language model), System 1 (the Motion Brain) and System 0 (the Interaction Brain); Sharpa counts System 0 and System 1 as CraftNet.
- Sharpa says it will share updates on CraftNet in phases.
Versions and notes in the key figures (5)
Model type
Note Sharpa's own term: "CraftNet is the combination of System 0 and System 1, forming a VTLA (Vision-Tactile-Language-Action) model that translates high-level, multimodal inputs into continuous, fine-grained actions executed by dexterous humanoid robots".
Controls
"System 1 generates coarse motions, including upper-body movements and pre-contact hand poses"; "System 0 produces human-level, fine-grained actions in the last millimeters of contact".
Inputs
Sharpa's model name spells out vision, tactile, language and action: "CraftNet is the combination of System 0 and System 1, forming a VTLA (Vision-Tactile-Language-Action) model that translates high-level, multimodal inputs into continuous, fine-grained actions executed by dexterous humanoid robots". Language comes from the layer above: "System 2 supplies semantic intent and/or explicit language instructions to System 1 within CraftNet". System 0 receives it "along with tactile and proprioceptive signals (e.g. force and torque)".
Access
Note "Sharpa will share updates on CraftNet in phases".
Base model
Note The reasoning layer uses an unnamed open-source model: "System 2 leverages an open-source VLM pre-trained on large-scale Internet data".
Sources
-
A
Primary source. www.
sharpa. · Retrievedcom/ blogs/ news/ sharpa-announces-craftnet-a-hierarchical-vtla-model-for-fine-manipulation -
B
www.
sharpa. · Retrievedcom/ blogs/ news/ sharpa-aims-to-manufacture-time-via-its-craftnet-vtla-model-and-stuns-ces-2026-with-live-autonomous-demos-for-fine-manipulation
One line per source, not per figure: a source has one retrieval date. The letter beside a figure points to the line it came from.