Built for Robotics Harness
The iNexBot embodied controller (the "cerebellum") is the real-time control layer of Robotics Harness: high-frequency state feedback + teleoperation data collection + force and admittance/impedance control — giving upper-level agents world state they can read, action outcomes they can judge, and safety boundaries they cannot cross, while translating model intent into stable, safe, engineering-ready real-robot motion.
iNexBot Embodied Controller
The brain understands; the cerebellum executes. Positioned beneath perception and decision-making yet above the actuators, the iNexBot embodied controller is the real-time control core that bridges up and down within Robotics Harness: it translates the large model's "intent" into stable, safe, engineering-ready joint motion, and faithfully hands the true state of the execution layer back to upper-level intelligence and training pipelines.
VLA / Large Models / Manipulation Learning
iNexBot
Servo / Joints / Bases
Hard Real-Time
Ubuntu + PREEMPT-RT real-time kernel with deterministic control cycles and low-jitter scheduling — a stable time base for high-bandwidth closed loops such as teleoperation and force control.
Multi-Platform
One runtime loads different JSON robot models, adapting to humanoids / 6–7 axis arms / SCARA / mobile bases — switching platforms means switching a model config.
Teleoperation Data Collection
High-frequency state feedback + teleoperation teaching produce real-robot datasets aligned with VLA / policy-learning formats, closing the "collection → training → deployment" data loop.
Force & Admittance/Impedance Control
Force-controlled joints + admittance/impedance control enable compliant interaction, contact-rich tasks, safe human-robot collaboration, and force guidance during collection — smoother teaching, steadier manipulation.
State Reading & Outcome Judgment
High-frequency feedback of joints, force, and execution progress with real-time outcome judgment — closing the two gaps physical-world agents struggle with most: reading world state and judging action results.
Safety Foundation
STO + emergency stop + safety PLC support, with layered limits on velocity, torque, and workspace — hard safety boundaries model output can never cross.
Making the controller part of the Harness
The success of coding agents proved a paradigm: what an agent achieves depends not on the model alone, but on the infrastructure around it — the harness. The physical world demands far more: a Physical AI harness must mediate at control, computing, and communication at once — reading world state, judging action outcomes, and bounding execution within safety limits. The iNexBot embodied controller exists for exactly this.
Reading World State
High-frequency feedback of joint position, force, and execution progress — agents always know what state the world is in.
Judging Action Outcomes
Execution monitoring and completion feedback give outcomes a factual basis — failures can be spotted and corrected.
Bounded Execution
Layered limits on velocity, torque, workspace, and safety zones — model output never crosses the line.
Execution Recording
Commands and true state recorded tick by tick — execution data that can be replayed, reviewed, and evolved.
Model-Agnostic
No binding to any VLA / large model / policy framework: intent in, motion out — swap the upper layer anytime.
Embodiment-Agnostic
One control interface across humanoids / 6–7 axis arms / SCARA / mobile bases — algorithm assets carry over.
The brain does not replace the cerebellum
VLA answers "what to do", but "how to stay stable, how to stay safe, how to scale to production" is the controller's battlefield — the foundation of our differentiated competition with pure-AI companies.
Neutral & Open
The Magna model: no allegiance to any robot-body vendor — we deliver control capability as an open module to robotics partners.
From Prototype to Production
Running in the lab doesn't count — the controller must hold up under line takt, safety certification, and long-term stability.
Deterministic Real-Time
Real robots don't wait: the hard real-time kernel keeps control cycles deterministic, so every intent from the model maps to predictable joint motion.
Data Loop
Collection, training, and deployment share one set of data formats and interfaces — the "collection → training → deployment" loop without manual data hauling.
Application Scenarios
Whole-body motion coordination, gait stability, and hybrid force-position control — a unified control foundation for bipedal and whole-body coordination.
High-frequency closed loop, trajectory tracking, and compliant control — up to contact-rich tasks such as assembly, insertion, and polishing.
Teleoperation teaching captures real-robot demonstration data at high frequency and fidelity, output aligned to VLA training formats.
Mobile base + robotic arm working in concert, with navigation and manipulation sharing one control and communication framework.
Open to Developers
Robot models are described in JSON; custom kinematics / dynamics extend via dynamic libraries (.so); SDK, examples, and docs work out of the box — minimizing integration cost so your team can focus on algorithms and hardware.
The Next Step for the Embodied Controller
Evolving continuously around Robotics Harness — more morphologies, richer compliance, and a more complete toolchain, to make the "collection → training → deployment" loop spin faster.
Platform Model Library
Continuously expanding JSON models and kinematics / dynamics libraries for humanoids / collaborative arms / SCARA / mobile manipulation.
Compliance & Force Control
Admittance / impedance control keeps evolving for more complex contact-rich tasks and human-robot collaboration scenarios.
Toolchain
SDK, examples, docs, and data tooling keep improving, shortening the path from algorithm prototype to real-robot validation.
Safety & Production
Continued investment in safety certification and production consistency — a controller that stands up to the production line and to time.
Bring Robotics Harness to your robots
Contact our sales team for embodied controller solutions, integration assessment, and live demonstrations.
Contact Sales