Does π0.5 run on SO-101?
Capabilities
| Capability | State | Notes | Evidence |
|---|---|---|---|
| Train / fine-tune | works | ev-hub-00ri-pi05-so101, ev-hub-tinkhireeva-pi05-so101, ev-hub-zz4321-pi05-so101, ev-card-dongyoonkim-pi05-so101, ev-card-zkxuerb-pi05-so101 | |
| Real-robot inference | works with caveats | About 90% observed success (trial count not recorded) on one task; a second checkpoint has phone-recorded live-evaluation videos with remote A100 inference (~1.3 s round-trip per 50-action chunk). | ev-card-dongyoonkim-pi05-so101, ev-card-zkxuerb-pi05-so101 |
| Simulation inference | no verified evidence | — | |
| Teleop + record | no verified evidence | — |
Caveats
- Community-reported only; not documented upstream or maintainer-tested.
- The quantified report (about 90% on one pick-and-place task) did not record the number of trials and ran through VLASH's PI05Policy on lerobot 0.4.1, not stock lerobot-rollout.
- The second report (pi05_base action-expert-only fine-tune, 50 episodes) shows only qualitative live-rollout videos and states the policy is not yet robust across the full workspace.
- Both are single-arm SO-101 setups with task-specific fine-tuning and fixed camera placement.
Evidence 5 sources; the cell shows the highest tier
- T3 · community reportedmodel-cardsource 2026-08-04verified 2026-10-25 (today)lerobot 0.4.1vlash 22cbabfee0f57874987c75a35a7dac129e695db0torch 2.7.1+cu126
Card for a pi0.5 checkpoint fully fine-tuned from lerobot/pi05_base on SO-101/SO-100 data reports a closed-loop evaluation on a physical SO-101 follower (2 cameras, step 40000, run via VLASH on an RTX 3090 Ti): about 90% observed success on one task, number of trials not recorded.
“Closed-loop evaluation on physical hardware, 2026-06-26. This is a qualitative observation on one task and one setup, not a benchmark.”
https://huggingface.co/dongyoonkim/so101-pi05-base/blob/56325bafcca9ec3155b8bfdf655be7df17e7f09f/README.md - T3 · community reportedmodel-cardsource 2026-06-02verified 2026-10-25 (today)
Card for a pi0.5 action-expert-only fine-tune of lerobot/pi05_base (train_config.json dataset zkxuerb/so101_right_arm_pick_cube_v1, meta/info.json robot_type=so_follower, 50 episodes) embeds two live-evaluation videos on a physical SO-ARM101 follower; qualitative only, not robust across the full workspace.
“Both clips are real-time, recorded on a phone during live evaluation on the SO-ARM101 right follower. Inference was served from a Modal A100 endpoint with ~1.3 s round-trip per 50-action chunk.”
https://huggingface.co/zkxuerb/pi05_so101_pick_cube_v1_expert_only/blob/efe2df1b047fa127f8fb58fbb6a221ac0d46236f/README.md - T3 · community reportedhub-checkpointsource 2026-01-04verified 2026-10-24 (1d ago)
config.json type=pi05; train_config.json dataset.repo_id=whosricky/so101-megamix-v1, whose meta/info.json robot_type=so101_follower (400 episodes).
https://huggingface.co/00ri/so101_megamix_v1_pi05 - T3 · community reportedhub-checkpointsource 2025-10-31verified 2026-10-24 (1d ago)
config.json type=pi05; train_config.json dataset.repo_id=tinkhireeva/so101_pick_place_yellow_objects_10_locations, whose meta/info.json robot_type=so101_follower (150 episodes).
https://huggingface.co/tinkhireeva/pi05_so101_pick_place_yellow_objects_10_locations - T3 · community reportedhub-checkpointsource 2025-10-12verified 2026-10-24 (1d ago)
config.json type=pi05; train_config.json dataset.repo_id=zz4321/so101_cube, whose meta/info.json robot_type=so101_follower (50 episodes).
https://huggingface.co/zz4321/so101_pi05
π0.5 inference speed by hardware robot-independent · inference cost only
| Compute | Runtime | Precision | Cams | p50 / p95 ms | Inference | Platform | Source |
|---|---|---|---|---|---|---|---|
| RTX 4090 | llama.cpp | fp16 | 2 @224x224 | 51.7 / — | 19.34 Hz | —conditionsboard: NVIDIA RTX 4090 · ggml_cuda_f16_f32_out: 0 (README preset for RTX 4090) Inference cost only; not task success. Full VLM + action-expert pass (ViT + LLM + 10-step action expert), LIBERO inputs (agent-view + wrist at 224x224). From the EagleVLA-Edge README RTX 4090 table (P50), PR #6 configuration. The README's naive π0.5 baseline on this GPU: 162.9 ms (not filed as a separate run: the baseline precision is not stated). | T3 · community reported |
| RTX 4090 | llama.cpp | fp16 | 2 @224x224 | 58.5 / — | 17.09 Hz | —conditionsboard: NVIDIA RTX 4090 · ggml_cuda_f16_f32_out: 0 (README preset for RTX 4090) Inference cost only; not task success. Full VLM + action-expert pass (ViT + LLM + 10-step action expert), LIBERO inputs (agent-view + wrist at 224x224). From the EagleVLA-Edge README RTX 4090 table (P50). The README's naive π0.5 baseline on this GPU: 162.9 ms (not filed as a separate run: the baseline precision is not stated). | T3 · community reported |
| Jetson AGX Thor 128GB | llama.cpp | fp16 | 2 @224x224 | 132.8 / — | 7.53 Hz | —conditionsboard: Jetson Thor (paper/README: "Jetson Thor") Inference cost only; not task success. Full VLM + action-expert pass (ViT + LLM + 10-step action expert), LIBERO inputs (agent-view + wrist at 224x224). Later configuration from the EagleVLA-Edge README (PR #6); the README notes an accuracy caveat for the two-switch GEMM preset on Thor. Naive PyTorch π0.5 on the same board per the paper: 457.9 ms (2.18 Hz) (not filed as a separate run: the baseline precision is not stated). | T3 · community reported |
| Jetson AGX Thor 128GB | llama.cpp | fp16 | 2 @224x224 | 202.8 / — | 4.93 Hz | —conditionsboard: Jetson Thor (paper/README: "Jetson Thor") Inference cost only; not task success. Full VLM + action-expert pass (ViT + LLM + 10-step action expert), LIBERO inputs (agent-view + wrist at 224x224). Paper's 'control frequency' (10.25 Hz) is 1/reaction time with VLM-skipping scheduling; not executed actions/s. Board inferred to be the 128 GB T5000 class from the paper's 130 W max-power entry for Thor (T4000 is 40–70 W). Naive PyTorch π0.5 on the same board per the paper: 457.9 ms (2.18 Hz) (not filed as a separate run: the baseline precision is not stated). | T3 · community reported |
| Jetson AGX Orin 64GB | llama.cpp | fp16 | 2 @224x224 | 271.1 / — | 3.69 Hz | power_mode 50W (paper label "Orin-50W")conditionsboard: Jetson Orin, 'Orin-50W' power mode (AGX Orin 64GB inferred, see ev-orin-board-inferred) Inference cost only; not task success. Full VLM + action-expert pass (ViT + LLM + 10-step action expert), LIBERO inputs (agent-view + wrist at 224x224). Later configuration from the EagleVLA-Edge README (time cache + cuBLASLt GEMM tuning, PR #6); not in the paper. Naive PyTorch π0.5 on the same board per the paper: 1420.8 ms (0.70 Hz) (not filed as a separate run: the baseline precision is not stated). | T3 · community reported |
| Jetson AGX Orin 64GB | llama.cpp | fp16 | 2 @224x224 | 374.5 / — | 2.67 Hz | power_mode 50W (paper label "Orin-50W")conditionsboard: Jetson Orin, 'Orin-50W' power mode (AGX Orin 64GB inferred, see ev-orin-board-inferred) Inference cost only; not task success. Full VLM + action-expert pass (ViT + LLM + 10-step action expert), LIBERO inputs (agent-view + wrist at 224x224). Paper's 'control frequency' (6.93 Hz) is 1/reaction time with confidence-based scheduling that skips the VLM; it is not executed actions/s and is not recorded as control_rate_hz. Naive PyTorch π0.5 on the same board per the paper: 1420.8 ms (0.70 Hz) (not filed as a separate run: the baseline precision is not stated). | T3 · community reported |
| Jetson Orin Nano 8GB (Super) | TensorRT | fp16 | 2 @640x480 | 438.47 / 438.93 | 2.28 Hz | jetpack 7.2.1 · power_mode MAXN_SUPER, clocks pinned with jetson_clocks (GPU 1020 MHz, CPU 1728 MHz) · tensorrt 10.16.2.10conditionsboard: Jetson Orin Nano Engineering Reference Developer Kit Super (8 GB) · l4t: R39.2.1 · os: Ubuntu 24.04.4 LTS · python: 3.12.3 · desktop: running Inference cost only; not task success. 300 s run on synthetic 640x480 observations; 10 denoising steps; latency is runtime-only over a pre-materialized observation ring. Whole-board RAM in use 6.39 GB (unified memory, desktop running). Actions were gated against the stock policy (cosine >= 0.99999) before timing. | T3 · community reported |
| Jetson Orin Nano 8GB (Super) | TensorRT | fp16 | 3 @640x480 | 643.39 / 644.92 | 1.55 Hz | jetpack 7.2.1 · power_mode MAXN_SUPER, clocks pinned with jetson_clocks (GPU 1020 MHz, CPU 1728 MHz) · tensorrt 10.16.2.10conditionsboard: Jetson Orin Nano Engineering Reference Developer Kit Super (8 GB) · l4t: R39.2.1 · os: Ubuntu 24.04.4 LTS · python: 3.12.3 · desktop: running Inference cost only; not task success. 300 s run on synthetic 640x480 observations; 10 denoising steps; latency is runtime-only over a pre-materialized observation ring. Whole-board RAM in use 6.18 GB (unified memory, desktop running). Actions were gated against the stock policy (cosine >= 0.99999) before timing. | T3 · community reported |
Embed this cell
Badge for READMEs and model cards. It always shows the current state.
[](https://caniembody.pages.dev/compat/so101/pi05)<iframe src="https://caniembody.pages.dev/embed/so101/pi05" width="380" height="132" style="border:0" title="π0.5 on SO-101 | CanIEmbody"></iframe>