Layout

• 6 inputs (sensor information)
• 12 neurons in the first hidden layer
• 8 neurons in the second hidden layer
• 4 outputs (up, right, down, left)

8/22/2026

I was thinking (chatting with Claude) and I realized that 6 inputs maybe isn’t enough for a 2D grid world.

This is because, first of all, our 2D Grid simulation is very simple yet not enough for 6inputs. The goal is very simple, a person needs to go from point A to point B while avoiding obstacles. The previous 6inputs were 1. posx 2. posy 3. goal_dist 4. goal_angle 5. obstacle_dist 6. comms_alive (0 or 1). At first glace, it looked fine but it had a major flaw.

1. What happens if there are multiple obstacles?, 2. Is only distance fine?

SO IT WASN’T FINE

Now there are 9 inputs, 1. posx 2. posy 3. goal_dist 4. goal_angle 5. obs_dist_up 6. obs_dist_down 7. obs_dist_left 8. obs_dist_right 9. comms_alive (0 or 1).

So it’s going to be 9-16-8-4 Total parameter number (I have no idea how this number came up, I’m gonna study) 9x16+16 + 16x8+8 + 8x4+4 = 244+136+36 = 344

According to Claude, this would work totally fine on FPGA (I don’t know how, I’m gonna study this as well)