From 5ec0af62c62de6c6f7ae4dad498d9dbcf676b062 Mon Sep 17 00:00:00 2001 From: Alexander Soare Date: Fri, 17 May 2024 15:05:40 +0100 Subject: [PATCH] Explain why n_encoder_layers=1 (#193) --- lerobot/common/policies/act/configuration_act.py | 3 +++ lerobot/configs/policy/act.yaml | 3 +++ 2 files changed, 6 insertions(+) diff --git a/lerobot/common/policies/act/configuration_act.py b/lerobot/common/policies/act/configuration_act.py index cc072083..95374f4d 100644 --- a/lerobot/common/policies/act/configuration_act.py +++ b/lerobot/common/policies/act/configuration_act.py @@ -119,6 +119,9 @@ class ACTConfig: dim_feedforward: int = 3200 feedforward_activation: str = "relu" n_encoder_layers: int = 4 + # Note: Although the original ACT implementation has 7 for `n_decoder_layers`, there is a bug in the code + # that means only the first layer is used. Here we match the original implementation by setting this to 1. + # See this issue https://github.com/tonyzhaozh/act/issues/25#issue-2258740521. n_decoder_layers: int = 1 # VAE. use_vae: bool = True diff --git a/lerobot/configs/policy/act.yaml b/lerobot/configs/policy/act.yaml index f09e6a12..7a12dcc2 100644 --- a/lerobot/configs/policy/act.yaml +++ b/lerobot/configs/policy/act.yaml @@ -66,6 +66,9 @@ policy: dim_feedforward: 3200 feedforward_activation: relu n_encoder_layers: 4 + # Note: Although the original ACT implementation has 7 for `n_decoder_layers`, there is a bug in the code + # that means only the first layer is used. Here we match the original implementation by setting this to 1. + # See this issue https://github.com/tonyzhaozh/act/issues/25#issue-2258740521. n_decoder_layers: 1 # VAE. use_vae: true