perf: headless recording via on-demand render runner

[wlgys8]

Summary

• Switch the motrixsim headless renderer to the render service's on-demand runner: frames are rendered only when a sync arrives and GPU readbacks are drained in-frame, so nothing renders while idle and recording runs at full render speed (fixes perf: multi-env headless video recording is superlinearly slow (per-env render + blocking readback per frame) #37)
• Remove the RenderConfig.render_fps cap option — headless recording is now always unlimited-rate; the render loop no longer paces captures to a fixed tick
• Add scripts/bench_headless_capture.py, a standalone probe that times render submission, blocking readback, take, and encode separately (serial vs pipelined schedule)
• Bump motrixsim to 0.10.1.dev123478

Verification

• scripts/bench_headless_capture.py: capture+encode at 1280x720 ~100 fps, 256x256 ~200 fps, 1080p ~80 fps (batch 16, unlimited rate)
• End-to-end play.py record_video=true on cartpole and microduck-walk-flat: 10 s clip records in ~12 s wall including startup
• motrix_env_core/tests + motrix_env_motrixsim/tests: 173 passed

[Copilot]

🟡 Changes recommended

The PR introduces at least one correctness issue (interactive renderer also gets fps=None unintentionally) and the new benchmark mis-attributes pipelined timing phases, making its results misleading.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

This PR improves headless video recording performance for MotrixSim-backed environments by switching headless rendering to an on-demand runner (render only on sync) and updating dependencies/tools to measure and validate the new capture behavior.

Changes:

• Switch headless MotrixSimRenderer to construct RenderApp using the on-demand runner (fps=None) to avoid idle rendering and improve capture throughput.
• Add a standalone benchmark script to attribute headless capture time across submission, readback wait, pixel extraction, and encoding (serial vs pipelined schedule).
• Bump motrixsim to 0.10.1.dev123478 in both pyproject.toml and uv.lock, and update renderer tests for the new RenderApp constructor signature/behavior.

File summaries

File	Description
uv.lock	Updates the locked motrixsim/motrixsim-core versions to the new dev build.
scripts/bench_headless_capture.py	Adds a benchmark probe to measure headless capture/encode costs and compare serial vs pipelined capture schedules.
motrix_env_motrixsim/tests/test_renderer.py	Adjusts mocks/assertions to account for the RenderApp(..., fps=None) headless constructor and updated sync/capture expectations.
motrix_env_motrixsim/src/motrix_env_motrixsim/renderer.py	Switches headless renderer creation to the on-demand runner and documents the capture/readback synchronization shape.
motrix_env_motrixsim/pyproject.toml	Pins motrixsim to 0.10.1.dev123478.

Review details

• Files reviewed: 4/5 changed files
• Comments generated: 2
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.