# mit-han-lab/pear > PEAR is a post-training engine for agentic robotics: it trains reasoning models to control real robots via RL in simulation, with zero demonstrations. - Magnitude: 6.0 out of 10 — Early signal - Stars: 439 total · +142 stars measured 2026-11-16, 00:28–09:36 UTC, ≈ 305 by evening - Star trust: star growth looks organic - Category: AI agents · Language: Python · License: Apache-2.0 · Created: 2026-11-13 · Last push: 2026-11-16 - GitHub: https://github.com/mit-han-lab/pear · Page: https://gitnova.dev/en/r/mit-han-lab/pear ## Useful for - Train Qwen3-27B to control a Franka robot in the LIBERO simulator via GRPO - Deploy the model on an RTX 5090 via vLLM and connect it to a YAM or SO-101 robot - Evaluate a policy on LIBERO-Plus, RoboDojo and RoboLab-120 benchmarks ## How to get started ```sh git clone https://github.com/mit-han-lab/pear.git && cd pear export WORK_DIR= # the model, the environments and the runs go here bash scripts/install.sh libero_run # installs the model, LIBERO and the training environment ``` Copied as is from the “getting started” section of the README. GitNova has not run these commands: check them in the original before running: https://github.com/mit-han-lab/pear#readme ## Why it’s here - Star-counter measurements on 2026-11-16 (UTC), 00:28–09:36: 297 → 439 stars (+142). This is the change over that interval. - Estimated end-of-day forecast: about +305 stars, using observed gains and the previous day. - The repository is 3 days old and already has 439 stars. With less than two weeks of history, there's no usual pace to compare the spike against yet. - Top new repositories this week: #74. - About 15 forks a day — people are taking the code. ## Star trust Star growth looks organic. Star-trust labels are heuristics based on the repository’s behavior, not a check of every stargazer. ## Numbers - Forks: 32 - Issues and pull requests: 4 - Watchers: 1 - Average over the last week: 180 per day - Usual pace: too little history (under two weeks) - Stars in the last hour (measured): 17 ## Stars per day, last 4 days (oldest → newest, today is partial) 2026-11-13 … 2026-11-16: 63, 125, 225, 142 ## Magnitude by day, last 2 days 11-15 5.2, 11-16 6.0 ## Spotted in now - Top new repositories this week: #74 ## Similar by description 1. **apple-aiml-research/ml-halo** — 1.9 · Cooling · Hardware, IoT & robotics · Python · +11 stars measured 2026-11-16, 00:33–09:41 UTC, ≈ 22 by evening Code for the HALO paper: a pipeline that generates synthetic data, trains a motion generator and a G1 humanoid tracking policy, and deploys them in simulation or on the real robot. It targets robotics researchers working on perceptive… Full card: https://gitnova.dev/en/r/apple-aiml-research/ml-halo.md 2. **google-deepmind/mujoco** — 3.1 · Steady · Science & research · C++ · +57 stars measured 2026-11-16, 00:29–09:38 UTC, ≈ 126 by evening A general-purpose physics engine for simulating articulated structures with contacts, used in robotics, biomechanics, graphics and reinforcement learning. Provides a C API, Python bindings, a native GUI and a Unity plug-in. Full card: https://gitnova.dev/en/r/google-deepmind/mujoco.md 3. **Open-RUA/uni-robot-agent** — 1.4 · Cooling · AI agents · Python · +6 stars measured 2026-11-16, 00:35–09:43 UTC, ≈ 13 by evening An open-source harness for robot-use agents with unified tool interfaces, Codex and Claude Code runtimes, and simulation environments RoboDojo, RoboUse and RoboLab. It lets you run and debug robot agents with open models. Full card: https://gitnova.dev/en/r/Open-RUA/uni-robot-agent.md 4. **cerulion-inc/cerulion** — 0.4 · Quiet · Hardware, IoT & robotics · Rust · +0 stars measured 2026-11-16, 00:36–09:45 UTC, ≈ 0 by evening Open-source Rust robot runtime that records runs, replays failures against new code, and lets AI agents control real robots within enforced safety limits. ROS 2 compatible. Full card: https://gitnova.dev/en/r/cerulion-inc/cerulion.md 5. **fuyichen0725/blinkmimic** — 0.0 · Steady · Hardware, IoT & robotics · Python · Not enough measurements today BlinkMimic trains a single whole-body motion-tracking policy for the Unitree G1 across all 40 LAFAN1 clips in about 3 minutes on one RTX 4090, using a differentiable world model and mjlab/MuJoCo Warp. Full card: https://gitnova.dev/en/r/fuyichen0725/blinkmimic.md --- Magnitude (0–10) measures how fast and how unusually interest in a repository is growing right now. It is not a quality score. Days are UTC. “So far today” is a fact; “expected by the end of the day” is a forecast. Summaries and use cases are written by an LLM (DeepSeek V4.1 Flash) from the README and may be inaccurate: verify specific claims (benchmarks, speed, hardware) in the repository itself. Data as of 2026-11-16 09:47 UTC, updated every 30 minutes.