Automatic control of suspension itself is nothing new. Fox Live Valve, RockShox Flight Attendant and more recently, SR Suntour’s TACT suspension products have been automatically adjusting suspension damping, with varying levels of success, for a good number of years now. However, the programming behind the function of these products is relatively fixed. There is no scope for the rider to give the system feedback on its performance. It can’t “learn” what the rider’s preferences are.
In contrast, what Shimano describes in US Patent 11866114 B2, is a system that can automatically adjust suspension behavior and seatpost and saddle position on-the-fly, that can also be “trained” by the rider to perform optimally for a given track.
I will admit that this patent made for some of the driest reading I’ll (hopefully) do all year. Like, as dry as the Sahara desert in a year of particularly low rainfall. Fear not. Here, I have attempted to distill it into an ever so slightly more hydrated and digestible format.
Machine Learning for Automatic Bicycle Suspension Control
The setup comprises a comprehensive data acquisition system with a host of sensors measuring a lot of parameters relating to how the bike is being ridden, and over what type of terrain. Speed, cadence, torque, accelerations, tire pressure and brake usage are all measured, as well as the bike’s yaw, roll, and pitch. Also in there are accelerometers on the suspension, recording information about how well the suspension is absorbing forces coming up from the wheel. There’s also a camera up front.
All that information is fed into a control unit, which is itself connected to a variety of electric motors or solenoid valves that are able to adjust things like spring rate, damper position, stroke length, lock-out, seat post height, as well as saddle position.
It’s not so difficult to imagine how a piece of software could be programmed to actuate changes in suspension settings to ensure that the bike’s behavior is appropriate to the terrain at any given moment. Though they don’t use all of the sensors mentioned here, Fox Live Valve and RockShox Flight Attendant suspension components operate on that basis. What’s unique to Shimano’s patent, is that the program can learn and alter itself accordingly as it collects more and more data – data that would be considered “training data” in AI and machine learning circles. And, some of that data can be in the form of direct feedback from the rider.
Fig.11 shows a selection screen, wherein the rider is asked to provide feedback on suspension settings and seatpost height changes that were implemented within the last 10 seconds of riding. The rider can select “Like” or “Dislike” (rendered somewhat hilariously in Facebook’s icons) to approve or disapprove of the change in question, feedback that then informs the learning module. This is a form of supervised learning where the system is able to take into account the rider’s preferences to make better decisions in future.
The document reads, “[the invention] achieves automatic control of the telescopic mechanism appropriate for the riding characteristics and the preference of the user being the rider of the human-powered vehicle.”
Fig.15 shows another screen stating which hypothetical courses have been learned by the program already – “Olympic Course” and “ABC Downhill”.
It’s entirely possibly my imagination is running a little wild here. But, is it possible that the aim is to develop a system for automatic suspension control that is incredibly specific to both the track being raced, and the rider’s preference?
Let’s think about how that might work over a World Cup XCO weekend. An athlete gets access to the course for 2-3 hours per day ahead of race day. They might get 6 or 7 full laps of the course, with the option to session the more technical sections of track where plethora line choice is on offer. On the first few laps, the rider uses the manual mode, and the learning module learns from the rider’s manual control of the suspension and seat post, while collecting information from the various sensors distributed about the bike. Thus, it will have a load of contextual information relating to when the rider decided to lock out their suspension, drop their post, etc., etc.
Later on, the rider would switch to the re-learning mode, wherein the system would take over operation of the seat post height and suspension adjustments. A short time after making each adjustment, the system would ask the rider for feedback on the change, with the simple “like” or “dislike”. That would provide the learning module with the rider-specific data needed to optimize the system further.
In patent speak, “automatic control of the telescopic mechanism suited for each of the travel courses depending on various situations or running environments in accordance with variety of input information can be achieved by training the learning model for each of the different traveling environments”.
Of course, the rider wouldn’t actually race with all of that hardware, owing to the weight penalty, but they could be happy enough to lug it around practice laps if it means their setup is optimized to the nth degree come race day.
But, Shimano Doesn’t Produce Suspension Components
This is the elephant in the room. While Shimano does manufacture the Koryak dropper seat post under the PRO Components brand, it does not produce suspension. So, what business have they patenting a machine learning method for automatic control of suspension adjustments?
It sure has left us pondering. I can’t really see this ever coming to market. It could simply be an R&D tool that Shimano is using to support their race teams. We reached out to Shimano for comment, to which they responded, “Shimano is constantly in development of new products but does not comment on rumors, innuendo, or speculation about products, whether they are in development or not”.
A Natural Evolution of the Dropper Seat Post?
While much of the patent contents focus on the implementation of a user-trainable machine learning program for the automatic adjustment of components, also tucked away in there is a description of what might be considered the next logical evolution of the dropper seat post.
It is not only able to telescope up and down to adjust saddle height, it can also adjust saddle tilt angle and fore-aft position on the rails by means of an electric motor. Such a dropper would allow a rider to dial in their saddle position while riding along, essentially allowing them to change their effective seat tube angle while riding.
It’s undeniable that the optimal combination for saddle height, tilt angle and fore-aft position for climbing is not the same as the optimal combination for descending. For a sustained uphill effort, the dropper will be at top-out, perhaps with the saddle nose angled down slightly to help the rider keep their weight forward, and possibly also shifted forward on the rails to keep the rider’s position biased forward to maintain a commanding position for pedaling. For descending, that saddle needs to be right out of the way; in addition to the obvious dropped position, it could also be beneficial to slide the saddle backwards on its rails with an altered tilt position. You’d essentially be looking the for the sweet spot wherein the saddle never interferes with your legs on a descent.
What Shimano has outlined could be loosely described as an electronically adjustable version of the Aenomaly Switchgrade. But, on top of the ability to adjustable saddle tilt, it can also adjust the fore-aft position.
Mention of such a dropper post doesn’t necessarily mean that Shimano is actually developing it. Sure, it could be our first glimpse of the next generation seat post from PRO. More likely, Shimano has added in those extra adjustable features to ensure their patent covers all bases, essentially future-proofing the scope of the patent.
