Nexus · Human Foundation Model

Nexus is the Human Foundation Model.

Mainstream world models predict the physical environment. Nexus predicts the person: their state, intent and boundaries. It is what makes an android able to live and work alongside people.

Human Foundation Model architecture

Nexus models the whole of humanity.

Built on a complete view of human health and potential, Nexus forms cross-modal, cross-context and long-horizon representations of people — a general foundation that can improve continuously.

Multimodal human state modelNEXUS / HFM

NexCap multimodal inputs

Nexa representing the embodied human data modelled by Nexus.

HFM

Unified human-state model

Temporal modelling and prediction engine

Spatiotemporal encoder plus state-space model, multimodal fusion Transformer, long-sequence prediction and uncertainty modelling.

Unified human-state representation

Pose / contact / load / balance · Emotion / arousal / fatigue / comfort · Intent / social state / cognitive state

Six core predictions

01

Body-state prediction

Pose, balance, load, motion trend

02

Affect-change prediction

Emotional intensity and arousal shift

03

Behaviour prediction

Next action and behavioural path

04

Intent inference

Approach, refusal, request, collaboration

05

Comfort and pain boundary

Comfort boundaries and pain thresholds

06

Social-response prediction

Social distance, acceptance, mode of reply

Data foundation

  1. 01NexCap full-body multimodal capture
  2. 02Human Biofidelity Dataset
  3. 03High-quality annotation and validation
  4. 04Nexus training and evaluation

Application loop

  1. 01

    Perceptual input

    NexCap real-time multimodal data

  2. 02

    World-model inference

    State understanding and future prediction

  3. 03

    Decision and generation

    Motion planning / language generation

  4. 04

    Android execution

    Natural interaction and motion output

  5. 05

    Feedback and learning

    Outcome evaluation and model update

Core value

Predictive interaction
From executing actions to predicting what comes next — the difference between reacting and understanding.
Safety
Predicting contact comfort and risk raises both safety and controllability.
Human likeness
Behaviour, affect and social response land closer to a real person.
Continuous evolution
A closed data loop keeps the model improving and individualising.

Turning multimodal human signals into one predictable, complete state — so machines can finally understand and anticipate people.

NexCap data foundation

The whole human state, on one map.

NexCap records 200+ synchronised channels across more than 60 modular modalities. Nexus aligns every signal to one clock and coordinate frame, then learns their relationships as a shared human-state representation.

Vision & motion
Optical mocap / depth / gaze / skeleton / IMU / velocity
Force & contact
Six-axis force / pressure array / force plate / torque
Muscle & neural
sEMG / activation / muscle synergy / neural stimulation
Physiology & affect
ECG / HRV / respiration / EDA / SpO₂ / temperature / arousal
Voice & audition
Speech features / microphone array / source localisation
Cognition & behaviour
Gaze / attention / cognitive load / response time / social context

Nexus is trained on data nobody else has

The model is only as good as the capture system behind it — which is why NexCap is a product and not an internal tool.

NexCapCommercial enquiries