Views
No views yet
1pip install -U huggingface_hub
2
3# Everything (~17 GB)
4hf download leekwoon/rcd-ogbench-checkpoints --local-dir ./rcd_checkpoints
5
6# Or a single environment, e.g. HumanoidMaze Medium only
7hf download leekwoon/rcd-ogbench-checkpoints \
8 --include "rcd_ogbench/logs/humanoidmaze-medium-stitch-v0/*" \
9 --local-dir ./rcd_checkpoints
10
11# Integrity check (optional)
12cd rcd_checkpoints && md5sum -c checksums.md5huggingface-cli download leekwoon/rcd-ogbench-checkpoints --local-dir ./rcd_checkpoints)| Environment | Run (planner) | Checkpoint | Size |
|---|---|---|---|
| pointmaze-giant-stitch-v0 | og_pntM_Gi_o2d_Cd_Stgl_..._T512 (U-Net planner) | state_1200000.pt | ~0.71 GB |
| antmaze-giant-stitch-v0 | og_antM_Gi_o2d_Cd_Stgl_..._T512 (U-Net planner) | state_1600000.pt | ~0.71 GB |
| antmaze-giant-stitch-v0 | og_antM_Gi_o29d_g2d_invdyn_h12 (inverse dynamics) | state_800000.pt | ~4 MB |
| humanoidmaze-{medium,large,giant}-stitch-v0 | og_humM_*_o2d_Cd_Stgl_..._T1000 | state_1800000.pt | ~0.77 GB each |
| humanoidmaze-{medium,large,giant}-stitch-v0 | og_humM_*_o69d_g2d_invdyn_h80_dm5_dout02 (inverse dynamics) | state_1600000.pt | ~35 MB each |
| antsoccer-{arena,medium}-stitch-v0 | og_antSoc_*_o17d_DiTd768_..._T512 (DiT planner) | state_1800000.pt | ~1.8 GB each |
| antsoccer-{arena,medium}-stitch-v0 | og_antSoc_*_o42d_g17d_invdyn_* (inverse dynamics) | state_1800000.pt | ~13 MB each |
| cube-{single,double,triple,quadruple}-play-v0 | og_cube*_DiTd1024dp12_..._T512 (DiT planner) | state_1600000.pt | ~2.5 GB each |
| cube-{single,double,triple,quadruple}-play-v0 | DQL low-level policy (epoch 200) | actor/critic/critic_target_200.pth | ~2 MB each |
args.json, model_config.txt,
and the *.pkl config files saved by the trainer.PointMaze uses a simple waypoint controller instead of a learned inverse-dynamics model, so only the planner checkpoint is provided for it (no*_invdyn_*run). AntMaze, HumanoidMaze, and AntSoccer each ship a planner plus a learned inverse-dynamics model. These planners are our own runs trained with the CompDiffuser-style codebase; loading a different (e.g. upstream CompDiffuser) checkpoint into these configs will not reproduce the RCD numbers, because the normalizer statistics and diffusion/overlap hyper-parameters are tied to the specific training run.
logs/ layout of the CompDiffuser-style codebase, so you can
merge it directly into your repository root:1# Locomotion (HumanoidMaze, AntSoccer)
2cp -r rcd_checkpoints/rcd_ogbench/logs/* <your_locomotion_repo>/logs/
3
4# Manipulation (Cube)
5cp -r rcd_checkpoints/rcd_ogbench_manipulation/logs/* <your_manipulation_repo>/logs/diffusion_epoch: 'latest', which resolves to the
included state_*.pt automatically. For example:1# HumanoidMaze Medium rollout
2python diffuser/ogb_task/ogb_maze_v1/plan_ogb_stgl_sml.py \
3 --config config/ogb_hum_maze/og_humM_Me_o2d_Cd_Stgl_PadBuf_Ft64_ts1k_h336_ovlp128_ovdm5_ts512_bs192_bd128_td96_drop02.py
4
5# AntSoccer Arena rollout
6python diffuser/ogb_task/ogb_maze_v1/plan_ogb_stgl_sml.py \
7 --config config/ogb_ant_soc/og_antSoc_Ar_o17d_DiTd768_PadBuf_Ft64_ts512_fs4_h160_ovlp56MditD384.py--dql_epoch 200) are included under
rcd_ogbench_manipulation/dql_results/<env>/.eval_cube_multi_mctd_rollout.py) expects them under
external/mctd_cube_public/dql/results/<run_name>/, where the original run name is<env>|exp|diffusion-ql|T-5|lr_decay|ms-offline|k-1|0|3|1.0|False|cql_antmaze|0.2|4.0|10| is not portable). Either restore that folder name,
or simply pass --dql_dir <path> to the eval script.