Physical-Intelligence/real-time-chunking-kinetix
Physical-Intelligence/real-time-chunking-kinetix has 636 GitHub stars and 47 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-19 23:08 UTC. Last push 2025-12-08.
Measured star growth
- in 24h
- 0
- in 7 days
- Collecting history
- in 15 days
- Collecting history
- in 1 month
- Collecting history
Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.
About
Simulated experiments for "Real-Time Execution of Action Chunking Flow Policies".
Developer ToolsREADME.md
OriginalSimulated experiments for the papers Real-Time Execution of Action Chunking Flow Policies and Training-Time Action Conditioning for Efficient Real-Time Chunking.
Installation
# Clone Kinetix submodule
git submodule update --init
# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh
# Install dependencies
uv sync
Pre-trained checkpoints and data
gs://rtc-assets/expert/ contains expert checkpoints generated by src/train_expert.py, and gs://rtc-assets/expert/data/ contains million-transition datasets for each level (generated by src/generate_data.py). Be aware that the expert/ directory is about 60GiB in total.
gs://rtc-assets/bc/ contains imitation learning policies for each level trained on the aforementioned data (generated by src/train_flow.py). These are directly usable with src/eval_flow.py.
Reproduce results
Note that, for all scripts, your number of GPUs must divide the number of levels (default 12) because computation is sharded over levels.
- Train expert policies:
uv run src/train_expert.py- By default, this will train 8 seeds per level for 65 million environment steps each.
- Checkpoints, videos, and stats are written to a wandb project called
rtc-kinetix-expertand the local directory./logs-expert/<wandb-run-name>. It is recommended to control other wandb options, like the run name, using environment variables.
- Generate data:
uv run src/generate_data.py --config.run-path ./logs-expert/<wandb-run-name>- For each level, this will automatically load the best-performing checkpoint for each seed (discarding seeds that didn't reach a certain success threshold).
- By default, 1 million environment steps are collected for each level using a mixture of expert policies.
- Data is written back to
./logs-expert/<wandb-run-name>/data/.
- Train imitation learning policies:
uv run src/train_flow.py --config.run-path ./logs-expert/<wandb-run-name>- This will load the data from step 2 and train flow matching policies for each level.
- Checkpoints, videos, and stats are written to a wandb project called
rtc-kinetix-bcand the local directory./logs-bc/<wandb-run-name>. It is recommended to control other wandb options, like the run name, using environment variables.
- Evaluate imitation learning policies:
uv run src/eval_flow.py --config.run-path ./logs-bc/<wandb-run-name> --output-dir <output-dir>- This will load the checkpoints from step 3 and evaluate them for 2048 trials per level by default.
- Currently, the script performs an exhaustive sweep over inference delay and execution horizon for all methods.
Training-Time RTC
To reproduce the results for training-time RTC, follow the following steps:
- Change
simulated_delayin the model config to 5. - Fine-tune the pre-trained checkpoint with simulated delay for 8 epochs:
uv run src/train_flow.py --config.run-path <run_path> --config.load-dir bc/24 --config.num-epochs 8wherebcis the contents ofgs://rtc-assets/bc/. - Evaluate as above.