Axonnic

Product Introduction

1. Definition: Axonnic is a human-to-robot skill transfer platform centered on a proprietary forearm EMG (electromyography) sleeve. It is a hardware-enabled AI data solution for robotics, specifically designed to capture and translate nuanced human dexterity into actionable training data for robots.
2. Core Value Proposition: Axonnic exists to solve the "data problem" in dexterous robotic automation. It enables manufacturers to automate complex, force-sensitive manual tasks—like sewing, delicate assembly, and tool handling—by recording the physical expertise of their best human workers. The core value is capturing the muscle activation and inferred effort that cameras miss, creating richer datasets to train more capable and adaptive robots.

Main Features

1. Forearm EMG Sleeve: A wearable device with 32-channel EMG sensors that non-invasively records electrical muscle signals from the forearm. It tracks individual finger movements and estimates fingertip force in real-time, without obstructing the hand. Technical validation shows an average joint-angle error of 4.58° against Apple Vision Pro and a processing speed of 50 predictions per second.
2. Multi-Modal Task Dataset Creation: The system doesn't rely on EMG alone. It synchronously captures raw EMG signals, video, and hand pose tracking (e.g., from a system like Apple Vision Pro) during real task performance. These streams are temporally aligned to build a comprehensive, task-specific dataset that correlates visual action with the underlying muscular "how."
3. Retargeted Policy Training & Deployment: Axonnic's in-house AI models are used to retarget human demonstrations into a robot's action space. The human EMG and motion data train a control policy for the robot. Critically, the deployed robot operates using only its own sensors (cameras, force-torque sensors); the sleeve is not used during runtime, ensuring the solution is scalable and non-intrusive in production.

Problems Solved

1. Pain Point: The inability to automate dexterous, non-repetitive tasks that involve soft materials, precise force control, or complex tool manipulation. Traditional robotic programming and vision-only imitation learning fail because they cannot capture the tactile intuition and force modulation of a skilled human.
2. Target Audience: Manufacturing engineers and automation leads in industries like apparel (sewing), automotive (cable harnesses, gasket placement), electronics (delicate assembly), and construction. Also, robotics and AI research labs seeking generalized datasets of human physical interaction for model training.
3. Use Cases: Force-sensitive sewing where fabric stretch and grip pressure are critical; delicate assembly of components where insertion feel matters more than precise path; machine tending and tool use that depends on how a tool is held and manipulated; field work documentation for skilled trades like electricians and plumbers to capture technique outside a controlled factory cage.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):

   • Pros: Captures pre-contact intent and muscle effort invisible to cameras. Leaves the hand completely free and unobstructed for natural work. Enables data collection in real-world, non-lab environments. Focuses on the hardest "last mile" automation problems. The robot operates independently post-training.
   • Cons: Requires an initial data collection phase with human experts. The accuracy of force estimation, while promising, is likely task- and individual-dependent. The system's performance is tied to the quality of the downstream robot policy training, a separate complex challenge. Adoption requires buy-in for a novel data-centric workflow.

2. Key Alternatives & Differentiation:

   • Vision-Only Imitation Learning (e.g., standard camera systems): Differentiation: Axonnic adds the crucial dimension of muscular effort and force intent, providing a denser, more explanatory signal for why an action is performed a certain way, leading to more robust policies.
   • Instrumented Gloves (e.g., sensor-laden data gloves): Differentiation: Axonnic's sleeve is less intrusive, allowing for completely natural hand movement and dexterity, essential for real tool use. It measures the control signal (muscle activation) further "upstream" than a glove which measures the output (finger bend).
   • Traditional Robotic Programming/Teleoperation: Differentiation: Axonnic aims to scale expertise capture. Instead of painstakingly programming one task or using complex teleop rigs, it records multiple human demonstrations to create a generalizable policy, potentially reducing setup time for new tasks.

Frequently Asked Questions (FAQ)

1. How does Axonnic's EMG sleeve work for robot training? The Axonnic sleeve records forearm muscle signals (EMG) while a human expert performs a task. This data, combined with video and hand pose, creates a rich training dataset. AI models use this to learn not just the movement, but the applied force and technique, generating a control policy for a robot to replicate the skill autonomously.
2. What is the difference between Axonnic and a motion capture glove? While both capture hand data, motion capture gloves track finger position. Axonnic's EMG sleeve measures the muscle activation signals that cause the movement and modulate force. This provides insight into the operator's intent and effort, and allows the hand to remain completely free of sensors, enabling more natural work.
3. Does the robot need to wear the Axonnic sleeve to operate? No. The Axonnic EMG sleeve is used only during the human demonstration and training phase. Once the robot's control policy is trained, it operates entirely using its own onboard sensors, such as cameras and force sensors, making the solution practical for continuous deployment.
4. What kind of tasks is Axonnic best suited for? Axonnic is specifically designed for dexterous, force-sensitive tasks that resist traditional automation. Prime use cases include sewing and handling deformable materials (fabric, cables), delicate assembly and finishing, skilled machine tending with tools, and documenting construction field work.
5. Who owns the data collected during an Axonnic pilot? According to Axonnic, data ownership terms are negotiated per engagement. Typically, recordings from a manufacturer's production line are used to automate that specific task. How the data may contribute to broader, generalized AI models is agreed upon contractually before the project begins.