COMPLETLY failed - Nothing went as planned with crowd simulation!

My failed attempt to realize crowd simulation using gradient descent algorithm. The idea here was to associate a cost to movement and obstacle avoidance in a way, that movement cost increases if the agent is not directly heading towards the goal, and obstacle cost increases if there is a higher chance of impending collision. We can then use gradient descent to descend upon these const functions to reduce them. Given that these cost functions are expressed in terms of speed and direction, as we descend we can obtain the speed and direction values that produce the least change in the cost function. It did not go as planned but here it is anyway. For this wondering - Va - Velocity of agent a - or agent under consideration Vaz, Vax - x & z components of velocity Va Pba - Position of other agent/obstacle WRT agent a The rest of the parameters are explained in detail in my previous crowd simulation videos.