From bdf02c9044891042bdf3bb0d14ed10d40d57ea39 Mon Sep 17 00:00:00 2001 From: shaoanlu Date: Sun, 28 Apr 2024 23:50:40 +0900 Subject: [PATCH] Update learning_note.md --- learning_note.md | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-) diff --git a/learning_note.md b/learning_note.md index f4a065f..22548ce 100644 --- a/learning_note.md +++ b/learning_note.md @@ -1,6 +1,7 @@ ## Learning note ### Main insight -- The model learns the residuals (velocity, gradient of the denoising) if possible. This greatly stabilizes the training. +- The model should learn the residuals (velocity, gradient of the denoising) if possible. This greatly stabilizes the training. +- Advantages of diffusion model: 1) capability of modeling multi-modality, 2) stable training, and 3) temporally output consistency. ### Scribbles - The trained policy does not 100% reach the goal without collision (there is no collision in its training data). - Unable to recover from OOD data. @@ -11,9 +12,11 @@ - For instance, it collides more on obstacles with a maximum radius of 1.5. - Collect more data to make everything interpolations instead of extrapolations. - Even though the loss curve appears saturated, the performance of the controller can still improve as training continues. - - The training loss curves of the diffusion model are extremely smooth btw. -- DDPM and DDIM samplers yield the best. +- The training loss curves of the diffusion model are extremely smooth btw. + - On the contrary, it might be difficult to know if the model is overfitting or not by looking at the trajectory as well as the the denoising process. +- DDPM and DDIM samplers yield the best result. - Inference is not in real-time. The controller is set to sun 100Hz. ### Things that didn't work -- Tried encoding distances to each obstacle. Did not observe improvement in terms of collision avoidance. \ No newline at end of file +- Tried encoding distances to each obstacle. Did not observe improvement in terms of collision avoidance. +- Tried using vision encoder to replace obstacle encoding. Didn't see performance improvement. \ No newline at end of file