HyperX AI Skill Trainer

Inspiration

HyperX is specifically known for its high-quality and affordable headsets, however, competition among keyboard and mice are quite high considering the fact that Razer, Corsair, and Steelseries exist. I wondered what HyperX could to gain a greater market share in the contemporary period besides hardware fixes, so I thought I could solve it using software! Competitive gaming is incredibly popular and as more people globally watch eSports, more average players want to get into the competitive edge. So, I thought of a project that could train players competitively while using the keyboard and mouse market of HyperX.

What it does

HyperX AI Skill Trainer is, as the name suggests, a skill trainer. The project uses the key inputs and mouse inputs of the user and compares it with the inputs of a skilled player (professional) and tries to point out the flaws in the user inputs as well as the methods the user could have used to gain an advantage. Since this is pertaining user-inputs, this would work for those playing a deterministic game as opposed to one that has RNG elements in it (e.g. Minecraft survival, CS:GO, Valorant, TrackMania, etc.) The application uses machine learning and artificial intelligence in order to locate which of your plays were game winning or game losing in each part of a specific scenario, making it quite helpful for those who want to learn alongside a pro player's inputs.

How it works

Step 1: Get competitive player inputs and get the user’s inputs (keyboard keystrokes and mice movement over the screen) Step 2: Using P-significance testing, find out whether a user’s inputs and the player’s inputs are statistically different in certain parts of the scenario. Step 3: Use machine learning and outcomes of a scenario to classify certain inputs as “game breaking” or “game winning,” while monitoring the inputs between the casual player and the competitive player. Step 4: Use machine learning to determine specific techniques specific to the scenario (A-D strafing, drifting: Shift + arrow keys) to highlight this to the casual player Step 5: Display it all in a UI!

Challenges we ran into

1. Random events and tiny input differences are difficult to predict. Deterministic scenario!
2. Getting player data for pro-players may be difficult as it takes 100s of pro-players and 100s of scenarios to get accurate/good data.
3. This is essentially a glorified keylogger, so extra security precautions must be taken.
4. If this product is to be created, a lot of hard code and logic must be behind input predictions.

What's next for HyperX AI Skill Trainer

Besides the following challenges that I thought of above, several improvements could be made. You could use better implementations to get user inputs. For example, HyperX could create a software similar to Nvidia ShadowPlay that could record the user's inputs in a specific game while recording the game itself (highlight), which can be compared along with a pro-player to see the differences between the inputs and the possible improvements in technique. Finally, this product could work with all games if the development team were willing to comply and send object/input information as well as RNG information to the application, granted that would make things extremely hard to code.

Built With

• c++
• hyperx
• machine-learning
• nspire
• python
• scipy
• tensorflow