TrainerRoad has today announced the launch of its all-new algorithm-based training plans, Adaptive Training.
Drawing from the company's database of millions of activities and workouts, Adaptive Training claims to take scientific training principles and apply a machine-learning algorithm to provide individualised coaching to each of its users.
Using a comparison of your power data vs the prescribed session, as well as a post-ride survey, Adaptive Training will analyse your performance in each workout in order to decipher how difficult you found it. It will then adjust the plan accordingly so that subsequent training progresses at an achievable, but optimal pace.
- Best power meters: Consistent power data for both indoor and real-world cycling
- Best turbo trainers: A roundup of the best-rated smart trainers
The theory and practice
TrainerRoad engineers have taken each workout from its thousands-strong database and associated them with one of the seven following energy systems (typically known as training zones): Endurance, Tempo, Sweet Spot, Threshold, VO2 Max, Anaerobic and Sprint.
They have then associated a progression score - ranging from zero to nine - to each workout, quantifying its difficulty.
For example, a 'Tunemah' is an over-under type threshold workout. It has three sets, each of which lasts 12 minutes, alternating between two minutes at 95% FTP and two minutes at 105% FTP. It has been given a progression level of 5.1. Meanwhile, 'McAdie +1' is a slightly harder version of the same type of over-under workout. This features four sets, each still lasts 12 minutes, but it alternates between one minute at 95% and two minutes at 105% FTP. This one has a progression score of 6.2. The understanding is that if you're unable to complete Tunemah, you'd also fail McAdie +1.
In a traditional, unadjusted plan, an athlete who failed a threshold workout in week one would still be fed progressively harder workouts in subsequent weeks and would most likely continue failing the workouts, resulting in sub-optimal training stress and a probable dent to motivation.
- What is FTP in cycling: How to perform your own test indoors
In this scenario, Adaptive Training would slow progression until a rider is able to complete Tunemah or an alternative threshold session with a similar score. Conversely, if a rider completed Tunemah with ease, Adaptive Training might skip past McAdie +1 in favour of an even more difficult alternative.
Users are then given a progression level for each of these energy systems based on success and failure rates of recent workouts, and TrainerRoad's 'Progression Levels' dashboard will give an overview of how well each of these individual energy systems is progressing.
In addition to calibrating your training after successful or failed workouts, the system can also account for skipped sessions or pre-planned time off. Working from its huge dataset, TrainerRoad can calculate reversibility (or how quickly your fitness drops when training stops) for each energy system. This will then adjust your plan so that when you return to training, you're tasked with an achievable but still-taxing amount of work.
Not following a plan?
The same technology has been applied to another new feature called TrainNow. This feature is designed for those who wish to add occasional training sessions to their otherwise unstructured riding in order to target specific improvements. TrainNow assesses a rider's current levels based on recent workouts and, once the rider has chosen their preferred season type (ie: Endurance, VO2 Max, etc) it suggests achievable but demanding workouts.
When is it available?
Adaptive Training will be rolled out in a closed beta from today while engineers iron out bugs. No timeframe has been set to this stage, but once the initial rollout has been completed, the software will be pushed to all users at no additional cost.
The launch will be announced in a live video on TrainerRoad's YouTube, which we've embedded below.
Thank you for signing up to Cycling News. You will receive a verification email shortly.
There was a problem. Please refresh the page and try again.