🕷️ AI Spider Training 2.0 | Live with Unity 6 & ML-Agents 22 | Ep. 3
15Episode 3: Refining Spider Model and Reward System
Welcome back to our AI spider training series! In Episode 2, we successfully set up our environment, copied code from examples, and ran an initial training session. Now, it’s time to refine our spider agents with some crucial updates.
In this episode, we’ll:
1. *Spider Model Refinements*:
*Add More Legs*: Update the spider model to have 4 legs with three joints each for realistic movement.
*Implement Realistic Raycast Sensors*: Integrate and configure multiple raycast sensors to act as the spider's eyes. These include:
*4 Anterior Median Eyes (AME)*: For detailed vision and depth perception, forward-facing.
*2 Posterior Median Eyes (PME)*: For a broader field of view, forward-facing but at a wider angle.
*1 Anterior Lateral Eye (ALE)*: On each side of the head, for peripheral vision.
*1 Posterior Lateral Eye (PLE)*: On each side towards the rear, for rear vision.
2. *Reward System Updates*:
*Set Minimum Reward Threshold*: Ensure a minimum reward threshold of -100 to manage performance more effectively.
*Adjust Ground Contact Penalty*: Change the die on ground contact penalty to give a -1 reward instead of an instant die penalty, promoting more exploration and learning.
3. *Curiosity and Memory Networks*:
*Implement Curiosity Networks*: Integrate a curiosity-driven reward system that encourages exploration by providing intrinsic rewards for discovering new areas or actions.
*Implement Memory Networks*: Integrate a memory module using RNN or LSTM networks to allow the spider to remember past experiences and use this information to inform future decisions.
Join us as we introduce these significant updates and push the boundaries of AI spider training. Whether you're following along or just curious, there's plenty to learn and explore.
Don’t forget to subscribe and hit the bell icon to stay updated with our progress. Let’s continue creating some dynamic, emergent behaviors together!
#unity6 #mlagents22 #aicoding #livecoding #spiderai #gamedevelopment #machinelearning #techstream