Sounds like a fun benchmark. Let’s start with a tight event loop, no garbage, and a spatial hash so the cat can query the light position in O(1). Keep the light’s trajectory deterministic so we can predict hits and pre‑compute a few frames ahead. That’ll let us hit real‑time and keep the chase alive without draining the GPU. Ready to dive into the math?

Sure thing. First, compute the light’s path as a parametric curve \(x(t)=\cos(\omega t), y(t)=\sin(\omega t)\). Sample it every millisecond, store the points in a uniform grid, and when the cat updates its velocity just grab the nearest cell’s light position. That keeps distance queries O(1) and the cat will sprint without hiccups. Ready to code?

import time, math, collections # grid parameters cell_size = 1.0 grid = collections.defaultdict(list) def add_light_to_grid(t): x = math.cos(0.01 * t) y = math.sin(0.01 * t) cell_x = int(x // cell_size) cell_y = int(y // cell_size) grid[(cell_x, cell_y)].append((x, y)) def nearest_light(cat_pos): cx, cy = int(cat_pos[0] // cell_size), int(cat_pos[1] // cell_size) candidates = [] for dx in (-1, 0, 1): for dy in (-1, 0, 1): candidates.extend(grid.get((cx+dx, cy+dy), [])) if not candidates: return cat_pos return min(candidates, key=lambda p: (p[0]-cat_pos[0])**2 + (p[1]-cat_pos[1])**2) cat_pos = [0.0, 0.0] while True: t = time.time() * 1000 # ms add_light_to_grid(t) target = nearest_light(cat_pos) # simple acceleration toward target dx, dy = target[0]-cat_pos[0], target[1]-cat_pos[1] dist = math.hypot(dx, dy) or 1 speed = 0.05 cat_pos[0] += speed * dx / dist cat_pos[1] += speed * dy / dist time.sleep(0.001) # 1 ms loop

Add a second cat, slower, and an obstacle that blocks cells. Just extend the grid logic: store obstacle cells and skip them when building the candidate list. Then give each cat its own speed multiplier. The obstacle will force the cats to reroute, giving more dynamic chasing. Ready to tweak the parameters?