When you watch someone reach for a coffee mug, you often know their intention before they even touch the handle. Are they going to take a sip, or are they moving the mug to clear space? Traditional philosophy suggested that intentions are "hidden" mental states, but a growing body of research shows that our "why" is physically encoded in the "how" of our movements.
This research direction focuses on kinematic signatures—the subtle, high-precision variations in velocity, grip aperture, and trajectory that reveal a person's goal during the very first moments of an action.
The Mechanism of Prospective Motor Control
The reason we can predict a goal before it is achieved is a phenomenon called prospective motor control. The human brain does not plan movements in isolated segments; instead, it pre-programs entire sequences. As Cristina Becchio and Caterina Ansuini argue in The visible face of intention: why kinematics matters, the ultimate goal of an action "leaks" into the initial phase.
For example, if you intend to pour from a bottle, your wrist height and finger spread during the initial reach will differ significantly from a reach intended to simply move the bottle. These variations are not random noise; they are a physical "surface flow" of intent. Atesh Koul and colleagues demonstrated in Decoding intentions from movement kinematics that machine learning models can use these "kinematic tells" to classify goals with high accuracy, even when the environmental context is removed.
The 200-Millisecond Window
How early can these intentions be detected? Surprisingly, the window is incredibly narrow. Aldo A. Faisal and Jovana J. Belić found that the intended goal of a manual action can be predicted within the first 200 milliseconds of movement initiation.
In The structured variability of finger coordination in daily tasks, they show that while the human hand has over 20 degrees of freedom, it actually operates within a low-dimensional subspace of just 4–5 dimensions. This redundancy allows the hand to begin adopting task-specific shapes—a process called pre-shaping—almost immediately. By the time a reach is only one-third complete, the kinematic signature is often distinct enough for a computer to "hallucinate" the final action.
Applications in Robotics and Prosthetics
This early predictability is the "holy grail" for assistive technology. For a prosthetic hand to feel natural, it must behave proactively rather than reactively. It needs to pre-shape its fingers during the reach, so it is ready to grasp the moment contact is made.
Researchers like Mehrshad Zandigohar and Mo Han have developed multimodal systems that combine muscle signals (EMG) with eye-tracking to achieve this. In Multimodal Fusion of EMG and Vision for Human Grasp Intent Inference in Prosthetic Hand Control, they demonstrate that fusing these streams allows a system to identify grasp intent over 700 milliseconds before physical contact. This lead time is critical for robotic hardware to actuate, enabling the fluid, synchronized movement seen in natural human-robot collaboration.
Similarly, Claudia Pérez-D'Arpino and Julie A. Shah have used these early kinematic cues to make collaborative robots "fluent" partners. By predicting a human's target within the first 400ms of a reach, a robot can begin its own task-related movement simultaneously, rather than waiting for the human to finish.
Clinical and Diagnostic Signatures
Beyond engineering, these kinematic patterns offer a window into neurodevelopment. In Prospective motor control obeys to idiosyncratic strategies in autism, Andrea Cavallo and colleagues used kinematic analysis to show that children with autism do plan their movements prospectively, but they use highly individualistic or "idiosyncratic" strategies.
While traditional clinical observations might miss these subtle patterns, machine learning can identify the unique "motor fingerprints" that distinguish different neurodevelopmental conditions. This suggests that the way we move isn't just about the task at hand—it is a reflection of how our internal cognitive models represent the world and our future actions within it.