So I’m trying to create a ragdoll that will learn how to balance and walk forward. In my scene I’ve made copies of the exact same prefab and spaced the ragdoll characters next to each other with a few meters of space in-between. But the issue is that when running the learning process, one of the ragdolls seem to develop a different behaviour than the other agents. If I remove that agent, another agent will be the one that develops the different behaviour.
With a different behaviour I mean that when all of the ragdolls seem to have developed a motion where they leans forward, the agent with the different behaviour will lean backwards and kick one of it’s legs forward, which none of the other agents do. This causes the reward to get stuck after a short while and not progress any further, which also makes sense that if two seperate ragdolls that are behaving differently are contributing to the same learning and reward then the results would be completely mixed together.
I have tried to remove all of the agents except one and duplicating it again multiple times, but the last one I duplicate will always be the one to develop a different movement/behaviour. It’s all seems very strange to me. Thank you for your time!