{"slug": "moving-from-ml-quant-work-into-robotics-a-concrete-first-week-measured", "title": "Moving from ML/quant work into robotics: a concrete first week, measured", "summary": "A developer transitioning from ML and quant work into robotics published a measured week-one log of reproducing a Soft Actor-Critic (SAC) baseline on MuJoCo's HalfCheetah using CPU-only PyTorch, reporting that each physics step takes roughly 50 microseconds while a SAC update at batch size 256 costs about 6 ms, making gradient steps rather than simulation the bottleneck. The work, scripted with an AI coding agent, includes per-run eval.csv logs, a plot script and a generated comparison table, released MIT-licensed in the sac-mujoco-repro repository, with a $1 packaged version offered to collect first reviews.", "body_md": "*Part of a public week-one log of moving from ML/quant work into robotics: what was actually run, what it cost, and the honest next steps. The scripting was done with an\nAI coding agent; every number below was measured on one CPU-only Linux box (Python 3.12,\nMuJoCo 4.0.1, PyTorch 2.15 CPU wheels).*\n\nThe robotics job market talks about VLA and foundation models, but the\n\ninterviews keep probing the layer underneath: state estimation, kinematics,\n\ncontact, and the discipline of measuring before claiming. MuJoCo is free, runs on\n\na laptop CPU, and every experiment below is checkable by a reader in under an\n\nhour.\n\n`pip install --target` + `PYTHONPATH` + CPU torch wheels. Cost: about an hour,\nmost of it reading error messages. The recipe is written down so it costs you\nten minutes.`env.step()` on HalfCheetah: ~50\nmicroseconds. So the bottleneck is gradient steps, not physics: a SAC update at\nbatch 256 costs about 6 ms. Multiply before you commit.\nEverything above ships with logs: `eval.csv` per run, a plot script, and a README\n\nwhose tables are generated. The claim is never \"I know robotics\"; the claim is\n\n\"here is a working baseline, here is what it cost, check it yourself\".\n\n*Everything above is checkable: code, full training logs and the generated\ncomparison table are MIT-licensed at [sac-mujoco-repro](https://gitlab.com/quilldatafix/sac-mujoco-repro). If you'd rather skip\nre-running the 1M-step training yourself, the packaged version — logs, learning\ncurves, checkpoints, README — is [on our store for $1](https://datafixtools.sell.app/product/sac-on-mujoco-working-cpu-reproduction-with-logs-and-comparison-table). The price is there\nto collect first reviews, not to gate the free repo.*", "url": "https://wpnews.pro/news/moving-from-ml-quant-work-into-robotics-a-concrete-first-week-measured", "canonical_source": "https://dev.to/quilldf/moving-from-mlquant-work-into-robotics-a-concrete-first-week-measured-bd3", "published_at": "2026-11-22 15:38:56+00:00", "updated_at": "2026-11-22 15:42:58.244809+00:00", "lang": "en", "topics": ["machine-learning", "ai-research", "ai-tools", "robotics"], "entities": ["MuJoCo", "PyTorch", "HalfCheetah", "Soft Actor-Critic", "sac-mujoco-repro"], "also_reported_by": [], "takeaway": "On a CPU-only MuJoCo setup, a HalfCheetah env.step() costs about 50 microseconds while a SAC update at batch 256 costs about 6 ms, so budget for gradient steps rather than physics simulation.", "recommendation": {"advice": "Profile step versus update cost before committing to a training budget: here physics steps ran ~50 microseconds against ~6 ms per SAC update at batch 256.", "playbook": "training-data", "playbook_url": "https://wpnews.pro/playbooks/training-data"}, "alternates": {"html": "https://wpnews.pro/news/moving-from-ml-quant-work-into-robotics-a-concrete-first-week-measured", "markdown": "https://wpnews.pro/news/moving-from-ml-quant-work-into-robotics-a-concrete-first-week-measured.md", "text": "https://wpnews.pro/news/moving-from-ml-quant-work-into-robotics-a-concrete-first-week-measured.txt", "jsonld": "https://wpnews.pro/news/moving-from-ml-quant-work-into-robotics-a-concrete-first-week-measured.jsonld"}}