Olympic Effort in Sports Performance
Pontem Performance x Team Houston Weightlifting x Aikynetix Collaboration
Currently in Paris (on way back from Cameroon amidst the “Blue Screen of Death” IT fiasco from last week…) and surrounded by over-priced Olympic souvenirs and banners featuring elite athletes about to make their lifetime run (or lift).
As we approach the Opening Ceremonies of the 2024 Summer Olympics (Paris), we wanted to share an update on our sports performance work in weightlifting, collaborating with Aikynetix. Most notably, this includes video analytics and technique support for Team Houston Weightlifting and Coach Tim Sword, one of Team USA’s certified weightlifting coaches and assistant Olympic coach in 2016 / 2020.
Previously on this topic, we shared some of the early stage work with Olympic hopeful Alyssa Ballard. She recently took 3rd place in the Junior Pan-American Games (Colombia), setting personal records in both the Clean & Jerk and Total. She will be competing next in the Junior World Championships (Leon, Spain). Congrats to Alyssa!
At that time, we were just starting the journey of how to incorporate Aikynetix’s analytics to assess lift performance, focusing on capturing primary angles and ensuring that the technology could be transferable across multiple biomechanical assessments. Recent updates have moved much further beyond this, so we are excited to share progress…
Team Houston Weightlifting x Aikynetix Collaboration
In January 2024, Pontem began our sponsorship of Team Houston Weightlifting. As part of our collaboration, Team Houston provided training videos from their athletes’ sessions of Olympic-style lifts such as the Snatch and Clean & Jerk. We worked with Aikynetix to see if their historically running-based platform could be adapted to other sports, such as weightlifting (as we had already with baseball, see bottom of post).
Given the Aikynetix focus on identifying proper form and enabling injury-free performance, Olympic weightlifting made for an interesting cross-collaboration opportunity. Particularly for those movements which are largely known and repeatable, the ability to track/trend across training sessions was appealing. While fractions of an inch make a difference in success/failure at the elite level, even weekend warriors engaging in Olympic weightlifting may benefit from a reduction in injuries that can come with bad form (i.e. see CrossFit…)
Similar to previous analytics for Alyssa, the Aikynetix team was able to do a much more detailed performance assessment for the Team Houston athletes (video below)
Aikynetix: Q&A
We discussed the process of training their tools to assess weightlifting videos, turning those into actionable insights, with Anton Galvas, Aikynetix CEO.
What was the main challenge you had in moving from running to weightlifting?
The primary challenge in transitioning our technology from running to weightlifting was the need to identify and analyze different key stages of complex weightlifting movements. Similar to analyzing running gait cycles, weightlifting required us to precisely capture and understand the dynamics of each phase of the lift to ensure accurate performance assessment and feedback, while at the same time addressing the occlusion problem.
What is the main challenge you still see remaining?
A significant remaining challenge is the development and refinement of 3D correction algorithms to enhance the accuracy of body joint angle assessments during weightlifting. We plan to implement video capture at a 45-degree angle, which will allow us to simultaneously assess body joint angles from both frontal and sagittal planes, reducing measurement errors and improving the comprehensiveness of our data.
We have talked offline that the process required to assess weightlifting has allowed you to better understand how to scale Aikynetix technology to multi-sport applications much quicker. Two part question: (a) how did you do that and (b) what does that mean for future sport applications?
(a) We leveraged the existing expertise within our team to rapidly develop algorithms capable of identifying key exercise stages and marking them in the kinematic profiles for joint angles across different sports. This was achieved by adapting our foundational technology to accommodate the unique demands and metrics of each sport.
(b) For future sports applications, this means we can swiftly extend our technology to new disciplines, providing interactive, AI-generated reports that not only evaluate performance but also recommend improvements. We are also exploring the possibility of advancing our models to include assessments of the center of mass (COM), which could lead to the estimation of ground reaction forces and other critical biomechanical metrics. The same approach was successfully used for running video analytics which we highly encourage to use it for the running performance improvement and injury prevention.
We have also discussed integrating coaching / feedback into the analytics assessments. Is this something that can be assisted with Generative AI?
Yes! We plan to utilize Generative AI to enhance the interpretability of our data outputs. By collaborating with subject matter experts (SMEs), we tailor AI-generated prompts to highlight crucial graphical and numerical insights from Aikynetix reports. This approach not only improves the user’s understanding of complex biomechanical data but also optimizes token usage, ensuring efficient and clear communication of analytical results. This will make our reports more accessible and actionable for coaches and athletes alike.
Converting “What You See” to Numbers
One of the biggest challenges faced in the use of video analytics-based training support is something that is encountered in any discipline: How do you get the knowledge out of our heads and down on paper? Or in a code? With the increase of remote athlete training, being able to screen multiple sessions quickly requires a more standardized approach. Three specific challenges emerged:
Without multiple (simultaneous) angles, how can a single camera capture the “3D world” that an in-person trainer can see?
Without physical measurements (i.e. accelerometers on the bar, force plates on the ground, etc.), are we losing valuable telemetry through the movement?
How do you translate ‘Coach-Speak’ into actionable insights that can be (a) standardized, (b) codified, and (c) replicated as part of a training assessment?
Similar to an experienced oil/gas operator in the field who knows how to “work a well” to get it flowing, what actual steps are they going through? While it may appear to be magic, or trial-and-error, there are indeed a list of physics-based “rules” that they are often working within. But, capturing that on paper - and translating that into a guidebook - can be very challenging. This is the elusive “Big Crew Change” dilemma of how to capture/retain knowledge and train (and scale) a new workforce.
As it relates to weightlifting, a similar challenge exists. Pontem’s own Fateh Sihota - a former world champion powerlifter / record holder and assistant coach for Team Houston Weightlifting - worked with Aikynetix to highlight the key “focus areas” that a coach generally looks for in the lifter.
Below is his list for the Clean & Jerk:
Symmetry at the beginning of the lift, foot position
Shoulders ahead of the bar?
Back Angle flat/arched
Does the bar move horizontally throughout the lift?
Full foot contact during pull phase
Speed of the bar off the floor, acceleration past the knee
Knees in or out through the lift
Do the arms remain long during the pull phase? Tow as opposed to pull.
Is a double knee bend performed?
Is a full extension realized at the top of the pull phase? On toes?
Jump force? Height of jump?
Does the lifter jump forward remain in place or jump back?
Speed of the lifter in racking the bar
Does the bar crash on the clavicle?
Elbows up and symmetrical, knees out?
Back position during squatting up
Symmetry standing with the bar racked
Dip phase – braced and knees out?
Jump – knees out? Jump force?
Bar straight up or forward?
Does the back foot land first?
Full extended shoulders, bar behind the ears?
Are knees inwards during jerk recovery?
Does the lifter walk the bar?
Stable for 3 seconds
The list is long, but a list of concerns that run through any subject matter expert (SME’s) mind trying to perfect their craft - weightlifting, restarting a well, cooking or painting - would look similar. These all have to work in concert, often in a “you’ll know it when you see it” sense. Which is great if you have access to the 20+ year veteran, but how can that be scaled? And, how can that be taught to a machine?
The challenge is then developing a useful tool is to somehow translate the above, putting some math into this list, and develop a tool that not only tracks performance, but starts to distinguish “good” from “bad”. Or at the Olympic level, “great” from “good”. The team at Aikynetix is working through this, as shown below in translating the video analytics into something that can be tracked, tweaked, and trended.
Being able to statistically show a complete lift is an important breakthrough. This is a work-in-progress that needs to be tested across different shapes/sizes of athletes and, ultimately, benchmarked against a success/fail criteria. There may be more than one way to get a successful outcome, but the goal is to understand ways to increase that chance of success and eliminate potential weak links to improve those changes. Additionally, one of the other focus areas is getting these Olympic-caliber athlete profiles loaded/stored/trained so they can be used by entry-level athletes as a means of comparing both visually and numerically. Everyone wants to see how they compare against a pro…so this is a chance to make that possible.
Behind the Video - Coming Soon…
Additionally, we have also been supplied with ancillary data from Team Houston for secondary effects that may impact what is seen on the video. Many of these could commonly be found in wearable devices, such as body weight and sleep. Others are more environmental, such as gym temperature and sequence of lift.
Being able to potentially couple the video analysis (the “how”) with this more ambient data (the “why”) offers potential advantages to understand not only how to optimize a given movement, but how to optimize a full training program, for world-class amateurs and weekend warriors alike. We may find that the data here is secondary (or tertiary) at best, but we shouldn’t rule anything out when looking for that proverbial edge.
Sports Performance - Baseball Recap (May ‘24)
Before we close, we wanted to recap our last sports analytics post from this summer - our baseball journey with Aikynetix:
In case you were wondering how the in-season video analytics and training efforts worked out for the Pontem kids when they got to Cooperstown:
11 Home Runs
Elusive “Back-to-Back-to-Back-to-Back” Home Run sequence (4 in a row…)
Note: this has only been done 8 times in major league baseball history, making it more rare than a no-hitter, perfect game, or almost any other achievement.
2nd Place finish in the Home Run Derby (losing to future MLB Hall of Famer Miguel Cabrera’s son, Christopher)
Lifetime of memories
Hard to draw too many conclusions about analytics’ role, although it definitely helped give life to instruction and helped “see” what we were continuously shouting. Ultimately, in the immortal words of Crash Davis (Bull Durham): “I don’t want to think about quantum physics. I don’t wanna think about nothin’. I just want to be…Don’t think. It can only hurt the ballclub”
And isn’t that just it: use advanced data/metrics relentlessly in training, so that when it comes game time, you can just perform.
