documentation_guidance.md

Classifer Guidance Tutorial

Create your own guidance function

1. Create diffusion_planner/model/guidance/<my_guidance>.py

def my_guidance_fn(x, t, cond, inputs) -> torch.Tensor:
    ...

    return reward

2. Add <my_guidance_fn> in diffusion_planner/model/guidance/guidance_wrapper.py

# diffusion_planner/model/guidance/guidance_wrapper.py

...

class GuidanceWrapper:
    def __init__(self):
        self._guidance_fns = [
            <my_guidance_1>,
            <my_guidance_2>,
            ...
            <my_guidance_N>
        ]

    def __call__(...):
        ...

...

3. Run sim_guidance_demo.sh
4. Enjoy.