×
How it works: Robot Belief
Based on "Robot Analysis Methods for Hide and Seek" (Goldhoorn 2017).
Visualizing Belief
Relative (Heatmap): Colors are normalized to the highest current probability. Red means "most likely spot compared to others". Useful for seeing diffusion.
Absolute (0-1): Strict probability scale. White is 0.0, Red is 1.0. Initially, the map will look white because probability is spread thin (e.g., 1/500 = 0.002).
Algorithms
1. Greedy (Frontier Exploration)
Robot marks map as Known/Unknown. Smart Mode acts as Frontier Exploration: it paths to the nearest unknown cell.
2. POMCP Belief (Grid)
Bayesian grid update. Smart Mode paths to the cell with the highest probability value ($P_{max}$).
init: Belief[x][y] = 1.0 / num_free_cells
loop:
Belief = Diffuse(Belief) // Prediction
for cell in visible: Belief[cell] = 0 // Observation
Normalize(Belief)
if Smart: MoveTo(Max(Belief))
3. Particle Filter
Sample-based belief. Particles are density-estimated into a probability grid [0.0, 1.0] for visualization. Smart Mode paths to the area with the highest density of particles.