Perceptual sensitivity supports online control, while perceptual errors drive motor memory formation during locomotor adaptation

To walk across a room, up a hill, or over uneven ground, the human body relies on a constant, invisible conversation between the brain and the muscles. This conversation has two distinct voices. One voice acts as a real-time referee, instantly correcting a stumble or adjusting a stride the moment a foot lands on a slippery patch. This is a reactive process, happening so fast it feels automatic and leaves no lasting trace once the obstacle is gone. The other voice is a learner. When the ground changes permanently—like walking on a new surface that demands a different rhythm—the brain builds a new memory of how to move. It updates its internal map of the world, storing a new pattern of movement that persists even after the person returns to normal ground. Scientists have long understood how this learning works in simple tasks, like moving a hand to catch a cursor on a screen, where the error is clear and controlled. But whether the same rules apply to the complex, full-body act of walking has remained a mystery, largely because it is difficult to tell what the brain "expects" to feel versus what it actually feels in such a dynamic environment.

A team of researchers at the University of Pittsburgh set out to solve this puzzle by watching how people learn to walk on a split-belt treadmill, a machine where the left and right belts move at different speeds. They wanted to see if the brain's ability to learn from mistakes in walking follows the same error-based rules seen in simpler tasks. To do this, they did something new: they asked the walkers to constantly report what they felt. While walking, participants were interrupted to answer a simple question: which leg felt slower? By tracking these answers alongside the actual muscle activity in their legs, the researchers could measure two different things. First, they measured "perceptual prediction errors," which is the gap between what the brain thinks is happening and what is actually happening. Second, they measured "perceptual sensitivity," or how sharp a person's sense of speed difference is, regardless of whether they are right or wrong about the direction.

The study involved fourteen young adults who walked on the treadmill for about an hour. The belts started at the same speed, then one sped up to 500 millimeters per second faster than the other, creating a lopsided rhythm. The participants had to adapt to this new, awkward feeling. The researchers found that the brain's learning process splits into two separate parts, driven by these two different perceptual abilities. The first part, the real-time corrections made during the early stages of walking on the fast belts, depended entirely on how sensitive a person was to speed differences. People who could detect small differences in speed between their legs made smaller, more efficient adjustments to their muscles right away. They wasted less energy and corrected their steps more precisely. However, this sharpness of feeling did not help them build a long-term memory of the new walking pattern.

The second part of the story, the formation of a lasting memory, was driven by something else entirely: the size of the prediction error. This is the mismatch between what the person expected to feel and what they actually felt. The researchers discovered that the people who felt the biggest difference between their expectation and reality ended up with the strongest memories of the new walking style. When the treadmill belts were returned to the same speed, these individuals showed the most dramatic "aftereffects," meaning their legs continued to move in the adapted pattern for a while, as if they were still on the split belts. Crucially, the size of this memory had nothing to do with how sensitive the person was to speed differences. A person could be very sharp at detecting speed changes but still have a weak memory if their internal expectation of "normal" walking didn't shift enough to create a large error signal.

This finding challenges the idea that a single mechanism drives all motor learning. Instead, it suggests that the brain uses a sharp sense of detail to fix problems as they happen, but it uses the magnitude of the surprise—the size of the error—to decide how much to update its long-term map of how to walk. The study also revealed that while the brain's internal reference for what feels "normal" shifted significantly to accommodate the split belts, the person's underlying ability to detect speed differences did not change at all. Their sensitivity remained steady, even as their sense of symmetry was recalibrated. This separation suggests that the brain can update its expectations about the world without altering the precision of its sensors.

The implications of this work extend beyond the treadmill. The researchers suggest that this dual-system view could help explain why different people struggle with walking in different ways. For instance, people who have had a stroke often have trouble making quick, real-time corrections but can still form long-term memories of new walking patterns. This fits the model perfectly: their ability to detect speed differences (sensitivity) might be impaired, hurting their immediate corrections, while their ability to learn from errors (prediction) remains intact. Conversely, older adults often show mixed results in walking studies; they might adapt their steady walking well but struggle to form strong memories of new patterns. This could mean their sensitivity is preserved, but the error-driven learning system is slowing down. By separating these two processes, scientists can now look for specific deficits in patients rather than treating walking adaptation as a single, monolithic skill.

The study did not find that the brain learns by simply optimizing for energy or stability in a general sense, as some alternative theories suggest. Instead, the data strongly supports the idea that specific prediction errors are the engine of long-term learning in walking. The researchers were careful to note that while the link between error size and memory strength was robust, the link between sensitivity and immediate correction was specific to the early phase of learning. As the body settled into the new rhythm, the influence of sensitivity faded, and the stored memory took over. This confirms that the brain switches gears, moving from a reactive mode driven by sensory sharpness to a predictive mode driven by the memory of past mistakes.

In the end, the research provides a clearer window into the mechanics of human movement. It shows that walking is not just a mechanical act of moving legs, but a sophisticated cognitive process where the brain constantly checks its expectations against reality. When the two do not match, the brain uses the size of that mismatch to rewrite its instructions for the future. The study confirms that error-based learning, a principle long known in the context of reaching for objects, is indeed the core driver of how we learn to walk, even in the complex, multi-sensory environment of a split-belt treadmill. The findings offer a new way to understand individual differences in movement and a potential roadmap for helping those who struggle to adapt to new physical challenges.