Skip to content

Support Train-Time RTC - #39

Draft
DavidHe137 wants to merge 11 commits into
mainfrom
train-time-rtc
Draft

Support Train-Time RTC#39
DavidHe137 wants to merge 11 commits into
mainfrom
train-time-rtc

Conversation

@DavidHe137

Copy link
Copy Markdown
Collaborator

No description provided.

DavidHe137 and others added 11 commits August 5, 2026 23:12
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Store rtc_prev_actions in client action space and route them back through
the input transforms, so DeltaActions re-anchors the frozen prefix to the
state of the observation being inferred, as during training. Bumps openpi
past the d15 configs and the make_example_actions removal.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Use an EMA of first_executed_index (ceiled) as the d param, falling back
to ceil(total_latency * control_hz) + 1 before the first ack. Realized
delay is preserved across resets like the latency tracker.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant