Canadian start-up smartARM is developing an AI-powered bionic arm designed to recognise everyday objects and automatically select an appropriate grip, potentially reducing the number of deliberate control steps required from the user.
The Toronto-based company’s vision-first approach combines a palm-mounted camera, multi-articulating fingers and its Dexterity OS software. Instead of asking the user to manually move between different grip modes, the system analyses the object being approached and determines how the prosthetic hand should respond.
A new Meta Newsroom feature also reveals how smartARM is experimenting with Meta’s open-source DINOv2 vision model and Meta AI glasses to provide the prosthesis with additional visual information from the user’s perspective.
The development illustrates how computer vision could become an increasingly important control layer in upper-limb prosthetics. However, smartARM remains in beta in Canada, and detailed specifications, pricing, regulatory status and independent clinical evidence have not yet been published.
Reducing manual grip selection
Multi-articulating prosthetic hands can offer several grip patterns, but accessing them may involve muscle signals, buttons, applications, gestures or sequential switching commands.
This creates a practical difference between the number of functions a device technically possesses and how easily those functions can be used during everyday activities.
Picking up a drinking glass and then changing to a spoon, for example, may require the user to consciously change modes before completing the second task. smartARM is attempting to move part of that decision-making process into the prosthesis itself.
The system’s onboard camera captures the object in front of the hand. Dexterity OS then interprets the visual information, identifies the object and selects a corresponding grip pattern before activating the fingers.
According to the smartARM website, the system currently recognises more than 365 everyday objects. The company says its object library can expand through software updates and become more personalised as the user interacts with the arm.
DINOv2 supports object recognition
smartARM is using DINOv2, an open-source computer-vision model developed by Meta, to help the prosthesis recognise objects from a small number of reference images.
Rather than requiring every possible object to be individually programmed through a lengthy development process, the model can analyse visual features and identify similarities between an object and previously learned examples.
Meta reports that smartARM can adapt to newly introduced objects using only a few photographs. Users can add objects through a connected smartphone application, allowing the recognition library to reflect the items that are important within their own home, workplace or daily routine.
This personalisation could be particularly useful because object design and daily activities vary considerably between countries and cultures. A prosthesis intended for international use must recognise more than a standard catalogue of North American household objects.
Meta AI glasses add a second viewpoint
The prosthetic hand’s palm-mounted camera provides a close view of the object, but its position may limit what it can see while the user is approaching or reaching.
smartARM is therefore exploring Meta AI glasses as an optional additional sensor. Using the Meta Wearables Device Access Toolkit, the glasses provide a first-person view from approximately the user’s eye level.
Combining the two perspectives could give the control system more contextual information about the object, the surrounding environment and the user’s apparent intention.
The glasses are presented as an optional addition rather than a requirement. This distinction will be important if the technology progresses towards clinical and commercial use, because relying on a separate consumer wearable could affect affordability, charging requirements, connectivity, privacy and long-term product support.
Clinical questions remain
Meta describes a bionic arm intended to work intuitively with little or no user training. That is a compelling ambition, but it should not be interpreted as evidence that professional assessment, fitting, occupational therapy or prosthetic training will become unnecessary.
Upper-limb prosthetic outcomes are influenced by socket comfort, suspension, electrode placement, limb presentation, control consistency, device weight, sensory feedback, task demands and the user’s personal objectives. Automated object recognition addresses one important challenge, but it does not remove these wider clinical considerations.
Future evaluation should examine:
- Object-recognition accuracy in different lighting and environments
- The consequences of selecting an unsuitable grip
- Task-completion time compared with conventional controls
- Performance with unfamiliar, damaged or partially obscured objects
- Battery life during continuous camera and AI operation
- Privacy and consent when wearable cameras capture other people
- Reliability without internet or cloud connectivity
- Ease of repair and software support
- User preference over extended periods
- Integration with established prosthetic fitting and rehabilitation pathways
The company currently indicates a battery duration of approximately 12 to 16 hours, depending on activity, but full technical specifications are expected closer to public launch.
Potential relevance across IMEA
For prosthetic services across India, the Middle East and Africa, automatic grip selection could reduce the cognitive burden associated with operating a multifunction hand. It may also make advanced prosthetic technology more approachable for users who find conventional mode-switching difficult or disruptive.
Accessibility, however, will depend on more than the AI model. Pricing, reimbursement, regulatory approval, component durability, local servicing and compatibility with different socket systems will determine whether the technology can move beyond selected demonstrations and beta users.
The use of open-source vision technology may support faster development, but the complete prosthesis remains a sophisticated combination of mechanical hardware, electronics, software and clinical care. Regional trials would therefore need to evaluate the device in hot climates, dusty environments, multilingual settings and healthcare systems with varying access to specialised upper-limb rehabilitation.
smartARM’s most important contribution may be its shift in control philosophy. Rather than asking users to continually tell the hand which grip to perform, the company is developing a prosthesis that interprets the task and prepares an appropriate response.
If that approach proves safe, reliable and clinically effective, vision-based control could become a significant addition to the future of upper-limb prosthetics.
- Meta: Canadian Start-up smartARM Uses AI to Create Intuitive Bionic Prosthetics
- smartARM Official Website
- smartARM Beta Programme
- Meta DINOv2 Open-Source Repository
- Meta Developer Platform