Toronto‑based startup smartARM is redefining what a modern prosthetic can be, building a bionic arm that doesn’t just respond to muscle signals—it sees the world. By integrating Meta’s DINOv2 vision model directly into the device, smartARM has created a prosthetic that can identify objects, understand their shape and context, and instantly choose the right grip without requiring manual switching or preset modes. It’s one of the clearest examples yet of how AI vision is transforming assistive technology from reactive tools into proactive, perception‑driven systems.
Image Courtesy : smartarm.ca
At the core of smartARM’s design is a camera embedded in the palm of the prosthetic. Instead of relying solely on EMG signals or user‑triggered grip patterns, the arm uses DINOv2 to analyze whatever the user is reaching for—a water bottle, a pen, a doorknob—and classify the object in real time. That classification is then mapped to a grip type, allowing the hand to adjust instantly. The result is a prosthetic that feels more intuitive and natural, reducing the cognitive load that many users experience when switching between grip modes on traditional bionic hands.
The company’s approach stands out because it treats vision as the primary input rather than an add‑on. Most prosthetics focus on improving motor control, but smartARM argues that perception is the missing piece. By giving the device the ability to understand its environment, the arm becomes more autonomous and more aligned with how biological limbs operate—combining sensory input with motor response. DINOv2’s ability to generalize across thousands of object categories without task‑specific training makes it ideal for this kind of real‑world prosthetic use, where unpredictability is the norm.
Beyond object recognition, smartARM is experimenting with contextual cues. The system can infer whether an object is fragile, heavy, or requires precision, adjusting grip force accordingly. This opens the door to prosthetics that not only grasp objects correctly but handle them safely and confidently. Early testers report that the arm feels “less robotic” and more like an extension of their own intent, because the device anticipates what they’re trying to do instead of waiting for explicit commands.
The startup’s work also highlights a broader shift in assistive technology: AI models originally built for general computer vision tasks are now being embedded directly into hardware. Meta’s DINOv2, known for its strong performance in self‑supervised learning, gives smartARM a foundation that doesn’t require massive custom datasets or constant retraining. As these models continue to improve, prosthetics could gain even richer environmental awareness—recognizing user habits, predicting actions, or adapting to new objects on the fly.
smartARM’s vision‑first prosthetic represents a major step toward more intelligent, autonomous assistive devices. By merging robotics, AI vision, and user‑centered design, the company is pushing prosthetics beyond mechanical function and into a future where artificial limbs can perceive, interpret, and respond to the world with human‑like intuition. As AI models grow more capable, this approach could become the new standard for bionic limbs, transforming daily life for millions of users who deserve technology that keeps up with their ambitions.