Sign in

Adrian Haith

@adrianhaith.bsky.social
575 followers 562 following 45 posts

Motor Control and Motor Learning

PostsRepliesMedia
Adrian Haith @adrianhaith.bsky.social · 15/07/2026
I don't know too much about balance specifically, but I can imagine it's similar to all other areas -- many different views that are hard to square with each other due to variable levels of description/detail and clashing terminologies. Can you link to your paper?
110
Adrian Haith @adrianhaith.bsky.social · 15/07/2026
I think this view broadly aligns with @adw.bsky.social and the Nguyen/Person proposal about the cerebellum doing “predictive” control without generating explicit *predictions* about future state
111
Adrian Haith @adrianhaith.bsky.social · 15/07/2026
You don’t need really that internal prediction at all. Information about the future state of the ball is already implicit in its current position + velocity. The fwd model computation doesn’t add any new information. Can just skip the prediction step and just use old observations to drive actions.
330
Adrian Haith @adrianhaith.bsky.social · 15/07/2026
There seems to be widespread acceptance that it is settled science that this kind of internal prediction of future states must exist in the motor system. I don’t really believe it though, and I think it’s caused a lot of confusion to have prematurely settled on this idea as a field
130
Adrian Haith @adrianhaith.bsky.social · 15/07/2026
Yup. Clashing terminology definitely doesn’t help. There’s no question motor behavior is anticipatory (perhaps could say “predictive”). Q is whether that capacity relies on an internal forward model that generates explicit *predictions* about the future.
120
Adrian Haith @adrianhaith.bsky.social · 14/07/2026
I’m not sure I fully understand the general principle though… Is it constructing a simple perceptual variable you can regulate to accomplish the task? Under what circumstances can you be sure it’s possible to derive something like that?
400
Adrian Haith @adrianhaith.bsky.social · 14/07/2026
Yes. It’s a great example for sure. I just worry it’s like solving a complicated equation where a lot of messy terms cancel out and you happen to get a clean analytical solution. But that doesn’t generally work and often you need a numerical solution. Not clear what the analog of that is though
210
Adrian Haith @adrianhaith.bsky.social · 14/07/2026
So how do you generalize the idea to cases like reaching out to catch a ball without moving your head?
210
Adrian Haith @adrianhaith.bsky.social · 14/07/2026
Okay. But, this does feel like a very specific problem where a simple direct perception solution happens to work. But what if I am sitting in a chair and toss me a ball? I will still move my hand to catch it but obviously I am not doing it by regulating the optical image of the ball.
220
Adrian Haith @adrianhaith.bsky.social · 14/07/2026
And I don't mean to pick on this paper too hard. It's actually a very thoughtful and thorough paper - much better than most other papers on this topic. It's just a shame that the alternative interpretation got swept under the rug and forgotten about, when I think it could be the correct theory.
020
Adrian Haith @adrianhaith.bsky.social · 14/07/2026
Yup. I generally agree with this view.
010
Adrian Haith @adrianhaith.bsky.social · 14/07/2026
That’s not the case for the Mehta and Schaal paper. The title + abstract state clearly that the motor system uses forward models, and it’s been cited >500 times, mostly for that fact. But then the small print admits it might not be forward models at all!
120
Adrian Haith @adrianhaith.bsky.social · 14/07/2026
Thanks!
000
Adrian Haith @adrianhaith.bsky.social · 14/07/2026
I’m skeptical they can really be dissociated behaviorally. It seems mostly a question of internal computation / control architecture which I think would be very hard or impossible to dissect behaviorally. I’d love to read that work though if you can point me to it
120
Adrian Haith @adrianhaith.bsky.social · 14/07/2026
I think so - if those explicit equations are about predicting future state from current state + motor command.
110
Adrian Haith @adrianhaith.bsky.social · 14/07/2026
So, yes, there are thousands of papers, but almost all already assume the idea of prediction/forward models as a given and then interpret their results through that lens.
240
Adrian Haith @adrianhaith.bsky.social · 14/07/2026
And I don't think neurophys evidence is much better. This recent review from Abby Person is v thoughtful on it from cerebellum perspective: www.nature.com/articles/s41.... Their conclusion is not too dissimilar from what Andrew W is advocating I think.
Ultimately, we suggest that the cerebellum may implement control through mechanisms that resemble internal models but involve model-free implicit mappings of high-dimensional sensorimotor contexts to motor output.
340
Adrian Haith @adrianhaith.bsky.social · 14/07/2026
I find the evidence for prediction to be pretty weak actually. One of the core behavioral papers people cite is Mehta & Schaal "Forward Models in Visuomotor Control": journals.physiology.org/doi/full/10..... But in the paper, they admit there is no decisive evidence for fwd models
352
Adrian Haith @adrianhaith.bsky.social · 10/07/2026
My favored approach is to always write the first pass of the code myself, then have the AI catch the things I forgot, fix syntax bugs, and take care of some more peripheral things I delegate to it. It's slower, but you keep full control of the project and maintain ability to read/write code.
010
Reposted by Adrian Haith
David Luque @davluque.bsky.social · 08/06/2026
New preprint! @martinezlopezp.bsky.social in this systematic review takes a look at the validity & reliability of current tools used for studying habits in the human lab. There are a lot of them! But studies about construct validity were sparse, while reliability analyses were almost nonexistent...
185
Reposted by Adrian Haith
Society for the Neural Control of Movement @ncmsociety.bsky.social · 04/02/2026
While preparing your #NCMKobe26 abstract, read through the highlights from #NCMPan25! The meeting highlight article is now available for review. Thank you to some of the scholarship winners from 2025 for putting the article together. journals.physiology....
055
Reposted by Adrian Haith
Society for the Neural Control of Movement @ncmsociety.bsky.social · 17/11/2025
Faculty at Predominantly Undergraduate Institutions: Your research matters and we want to see you at #NCMKobe26! NCM offers a fellowship to support PUI faculty presenting at the meeting. Up to $1,500 in travel support. More info: ncm-society.org/dive...
012
Reposted by Adrian Haith
Juan Gallego @juangallego.bsky.social · 13/11/2025
Come share your passion about motor control, sensory systems, neurophysiology, neurotechnology, and more at #NCMKobe26 !!
0153
Adrian Haith @adrianhaith.bsky.social · 03/11/2025
Deadline to submit for oral presentations at #NCMKobe26 is in just under a month! We especially welcome Team/Panel proposals: 2 hour session w 4 talks + discussion. Historically, the acceptance rate is higher for panels than for individual talks. Poster submission closes in February.
052
Adrian Haith @adrianhaith.bsky.social · 29/10/2025
Lastly, I should add -- I'm not the first to propose policy gradient RL for modeling human motor learning. Check out @nidhise.bsky.social 's excellent paper arguing for policy-gradient RL in locomotor learning:
031
Reposted by Adrian Haith
Nidhi Seethapathi @nidhise.bsky.social · 29/10/2025
Our 2024 paper showed that policy gradient RL (with performance-based memory updates) predicts long-horizon motor learning. Now, @adrianhaith.bsky.social shows that policy-gradient RL also explains learning in other shorter horizon tasks. Exciting! www.biorxiv.org/content/10.1...
biorxiv.org
Policy-Gradient Reinforcement Learning as a General Theory of Practice-Based Motor Skill Learning
Mastering any new skill requires extensive practice, but the computational principles underlying this learning are not clearly understood. Existing theories of motor learning can explain short-term ad...
1132
Adrian Haith @adrianhaith.bsky.social · 24/10/2025
I suspect we stand to learn more from the roboticists than the other way around. Progress there is accelerating rapidly! But maybe there will be insights that can go in the other direction too. I think it’s a very interesting direction to explore!
030
Adrian Haith @adrianhaith.bsky.social · 21/10/2025
Yes. A lot of these challenges are being addressed already in robotics, with basically PG methods. I believe that ultimately policy gradient + smart/adaptive curriculum/reward function can get you pretty far. That latter part is the real role of cognition in skill learning.
130
Adrian Haith @adrianhaith.bsky.social · 21/10/2025
I'm certainly not proposing this as a "Theory of Everything", but rather as an alternative foundation to error-based models of learning. Very feasible I hope to extend the theory in future to account for the kinds of things you mention. And some things may even make more sense from this perspective.
030
Adrian Haith @adrianhaith.bsky.social · 21/10/2025
As for contextual interference, savings etc. Those things are increasingly viewed as occurring at the level of retrieval and/or separation of policies across tasks, rather than low-level learning rules. In that case, it's quite compatible with the underlying policy learning rule being model-free RL
140
Adrian Haith @adrianhaith.bsky.social · 21/10/2025
Offline learning could be easily explained through replay - RL applications in robotics etc. do exactly this, and there's plenty of evidence something like this occurs during sleep. So it's very compatible with that.
110
Adrian Haith @adrianhaith.bsky.social · 21/10/2025
Thanks, JJ. It doesn't directly predict those things. In our experiments, we actually haven't found there to be much forgetting of skills you learn through practice. People retain pretty much everything in our de novo task, even after a year: drive.google.com/file/d/11M0l...
110
Adrian Haith @adrianhaith.bsky.social · 21/10/2025
Thanks, JT!
010
Adrian Haith @adrianhaith.bsky.social · 20/10/2025
Please check out the paper for more details and hopefully an accessible intro to policy-gradient RL if you’re not familiar with it. I welcome any feedback. If you’d like to dabble with the models, code for all simulations is available at: github.com/adrianhaith/PolicyGradientSkillLearning /End.
380
Adrian Haith @adrianhaith.bsky.social · 20/10/2025
I’m excited about the potential of this approach. Progress on understanding the kinds of motor learning that really matters for sports, rehab, and development, has been pretty limited in recent decades. I’m hopeful that a concrete and simple computational theory can help spur progress.
130
Adrian Haith @adrianhaith.bsky.social · 20/10/2025
And in a precision movement task, requiring precise, speeded movements through an arc-shaped channel (Shmuelof, Krakauer & Mazzoni, 2012):
330
Adrian Haith @adrianhaith.bsky.social · 20/10/2025
In a cursor-control task with a highly non-intuitive mapping (often described as “de novo” learning, since it requires learning a brand new controller rather than adapting an existing one; Haith, Yang, Pakpoor & Kita, 2002):
130
Adrian Haith @adrianhaith.bsky.social · 20/10/2025
Across three quite different tasks, policy-gradient RL models of learning account very well for the patterns of improvement in the mean and variance of people’s actions across a range of tasks. In Müller and Sternad’s skittles task (Sternad, Huber, Kuznetzov, 2014):
140
Adrian Haith @adrianhaith.bsky.social · 20/10/2025
Policy-gradient RL is a simple, model-free RL method that is a pillar of impressive recent advances in robotics. Here, I show that a trial-by-trial learning rule based on policy-gradient RL accounts remarkably well for the way people improve at a skill through practice.
140
Adrian Haith @adrianhaith.bsky.social · 20/10/2025
New Pre-Print: www.biorxiv.org/cgi/content/... We’re all familiar with having to practice a new skill to get better at it, but what really happens during practice? The answer, I propose, is reinforcement learning - specifically policy-gradient reinforcement learning. Overview 🧵 below...
biorxiv.org
Policy-Gradient Reinforcement Learning as a General Theory of Practice-Based Motor Skill Learning
Mastering any new skill requires extensive practice, but the computational principles underlying this learning are not clearly understood. Existing theories of motor learning can explain short-term ad...
36422
Adrian Haith @adrianhaith.bsky.social · 12/09/2025
New pre-print! We attempt to survey the two different universes of motor learning research: basic (meetings like NCM, MLMC) and applied (e.g. NASPSA), and consider what these fields can learn from each other and what the future might look like if they can be better integrated. More in Eric's 🧵 👇
osf.io
OSF
0135
Adrian Haith @adrianhaith.bsky.social · 06/09/2025
I agree there are other cons that have a more subtle impact and will be harder to mitigate - like failure to properly cite or attribute ideas, or potentially leading everyone up the same path. Those are the ones to be concerned about, not so much AI hallucination/confabulation.
000
Adrian Haith @adrianhaith.bsky.social · 06/09/2025
I would say a “bad” AI user is someone who uses it without being wary of its limitations and without properly validating its output. I expect most scientists to be capable of using AI with appropriate skepticism of its output.
200
Adrian Haith @adrianhaith.bsky.social · 05/09/2025
Bad, lazy scientists are nothing new, but they are a small minority. I'm confident most of us will be able to use it wisely to make our science better
100
Adrian Haith @adrianhaith.bsky.social · 05/09/2025
So many pros. Most 'cons' are avoidable by basic common sense: don't just blindly assume that what it outputs is correct or true. The alarmism seems to be all about how *other* people will use it - bad, lazy scientists using it to do bad, lazy science.
100
Reposted by Adrian Haith
Nature Reviews Neuroscience @natrevneuro.nature.com · 18/06/2025
Cerebellar circuit computations for predictive motor control — a Review by Katrina P. Nguyen & Abigail L. Person www.nature.com/articles/s41... #neuroscience #neuroskyence
nature.com
Cerebellar circuit computations for predictive motor control - Nature Reviews Neuroscience
The cerebellum helps ensure the speed and accuracy of movements, but its precise contributions to movement control are unclear. Nguyen and Person here evaluate evidence for and against feedforward mot...
1239
Reposted by Adrian Haith
Yue Du @yuedu.bsky.social · 03/06/2025
Very excited to share our new paper with @adrianhaith.bsky.social, now published in @nathumbehav.nature.com.
1279
Reposted by Adrian Haith
Nature Human Behaviour @nathumbehav.nature.com · 03/06/2025
In this article, @jetrach.bsky.social and McDougle show that motor responses can form part of structured, graph-like memory representations. @actlab.bsky.social www.nature.com/articles/s41...
nature.com
Mental graphs structure the storage and retrieval of visuomotor associations - Nature Human Behaviour
Trach and McDougle show that motor responses can form part of structured, graph-like memory representations.
2267
Reposted by Adrian Haith
Nature Human Behaviour @nathumbehav.nature.com · 03/06/2025
In this article, @yuedu.bsky.social and @adrianhaith.bsky.social show that behavior can become habitual in two different ways, involving response initiation and response preparation, respectively www.nature.com/articles/s41...
nature.com
Dissociable habits of response preparation versus response initiation - Nature Human Behaviour
Du and Haith show that behaviour can become habitual in two different ways, involving response initiation and response preparation.
12613
Adrian Haith @adrianhaith.bsky.social · 25/04/2025
I recommend this paper: www.frontiersin.org/journals/com...
frontiersin.org
Frontiers | Are muscle synergies useful for neural control?
The observation that the activity of multiple muscles can be well approximated by a few linear synergies is viewed by some as a sign that such low-dimensiona...
110