AI scientist Muse beats Bayesian optimization on battery fast-charging search in all three benchmarks
YashasGunderia · x · 2026-11-05
The team behind Translucent, a physical-test infrastructure for hardware teams, benchmarked their AI scientist Muse on battery fast-charging protocol search.
- Task: find the fastest 10–80% charging protocol for a simulated LG M50 cell within temperature and voltage constraints, with a budget of 8 simulated experiments per method
- Results: Muse landed on the best known charge in all three cases, beating a constrained Bayesian-optimization baseline by 5–12 minutes of simulated charging time; in the 15°C held-out extrapolation case it hit the best known median (34.13 min)
- Muse also adapted mid-run when the anode-potential limit was raised from +10mV to +25mV, rather than following a fixed experiment list
- Caveat: results are computational, under a frozen PyBaMM model, not bench measurements
More from Research
- ICML Weight-Space Symmetries keynote slides survey weight-space learning — orvieto_antonio · 2026-11-05
- PhD researcher launches 'paper nostalgia' series with Real NVP coupling-layer explainer — LucaAmb · 2026-11-05
- New theory says curricula help with unlearning simple rules, not learning — summerfieldlab · 2026-11-05
- uni-robot-agent gets Qwen working in sim and on real robots, launching OpenRUA — teortaxesTex · 2026-11-05
- Fabricated references in 500K-paper sample surge from 6.3 to 23.0 per 10,000 since 2023 — skdh · 2026-11-05
- Nature paper warns recursive training on AI-generated data causes irreversible model collapse — solyarisoftware · 2026-11-05