Game Programming

Game Programming

Why it took years to teach the robots in Arc Raiders to walk

The legged robots in Arc Raiders learned to walk through reinforcement learning, and getting there took five to ten people several years. Following the GDC and NDC talks and interviews of Embark's machine learning lead, I go through the choice of physical robots, balancing rewards, the scared robot and the two brains, AMP bringing the animators back in, the robots' eyes, and the limits of this approach.

Nov 8, 2026
Game Programming

Understanding the fixed-timestep loop, properly this time

The same jump code reaches 1.47 m at 30 fps and 1.57 m at 144 fps. The standard fix for this problem, the fixed-timestep loop, collects the real time that passed in a tank and only ever scoops it out in pieces of the same size (dt), running physics once per piece. Whatever is too small to scoop carries over to the next frame, and the screen draws a blend of the previous state and the current one, weighted by that leftover.

Nov 8, 2026