[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-82174":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":14,"subscribersCount":14,"size":14,"stars1d":15,"stars7d":16,"stars30d":16,"stars90d":14,"forks30d":14,"starsTrendScore":17,"compositeScore":18,"rankGlobal":9,"rankLanguage":9,"license":19,"archived":20,"fork":20,"defaultBranch":21,"hasWiki":22,"hasPages":20,"topics":23,"createdAt":9,"pushedAt":9,"updatedAt":24,"readmeContent":25,"aiSummary":26,"trendingCount":14,"starSnapshotCount":14,"syncStatus":15,"lastSyncTime":27,"discoverSource":28},82174,"SonicStar","BlackOtters\u002FSonicStar","BlackOtters","Open-source Unitree G1 Vision-Language-Action stack for teleop data collection, SonicLatent training, simulation, and real-time whole-body policy deployment(real world deployment TBD).",null,"Python",33,3,1,0,2,5,6,1.81,"MIT License",false,"main",true,[],"2026-06-12 02:04:23","# SonicStar\n\nUnitree G1 的 VLA 开源仓库，分成两块：\n\n- `starVLA\u002F`: 训练、数据集、推理部署\n- `wbc\u002F`: 部署、遥操作采集、仿真\n\n详细背景和通用流程可直接看：\n\n- https:\u002F\u002Fnvlabs.github.io\u002FGR00T-WholeBodyControl\u002Ftutorials\u002F\n- https:\u002F\u002Fstarvla.github.io\u002Fdocs\u002Fzh-cn\u002F\n\n## 采集数据\n\n在 `wbc\u002F` 下启动采集链路,在不同终端下依次运行（建议把https:\u002F\u002Fgithub.com\u002FNVlabs\u002FGR00T-WholeBodyControl.git 克隆下来，在那个仓库下跑，本仓库仅供参考示意）：\n\n```bash\npython gear_sonic\u002Fscripts\u002Frun_sim_loop.py --enable-image-publish --enable-offscreen --camera-port 5555\npython gear_sonic\u002Fscripts\u002Frun_camera_viewer.py --camera-host localhost --camera-port 5555\nbash deploy.sh --input-type zmq_manager sim\npython gear_sonic\u002Fscripts\u002Frun_data_exporter.py --task-prompt \"pick up the cylinder and throw it into the trash bin\"\npython gear_sonic\u002Fscripts\u002Fpico_manager_thread_server.py --manager\n```\n\n采集完毕后根据starVLA教程https:\u002F\u002Fstarvla.github.io\u002Fdocs\u002Fzh-cn\u002Ftraining\u002Flerobot-dataset\u002F 修改数据集的meta\u002Fmodality.json\n也可以参考我自己采集的数据集https:\u002F\u002Fhuggingface.co\u002Fdatasets\u002FTang-keke\u002Fmerged_dataset_001\n注意我的数据集中meta\u002F的source_episodes、stats_gr00t、steps_data_index不是必要的文件，不需要参考\n\n## 训练 VLA\n\n```bash\nbash examples\u002FSonicLatent\u002Ftrain_files\u002Frun_sonic_latent_train.sh\n```\n\n默认配置：\n\n```bash\nexamples\u002FSonicLatent\u002Ftrain_files\u002Ftrain_sonic_latent.yaml\n```\n\n我的训练数据集从`GR00T-WholeBodyControl\u002F` 也就是`wbc\u002F` 采集后，放在starVLA\u002Fplayground\u002FDatasets\u002F里面（建议把https:\u002F\u002Fgithub.com\u002FstarVLA\u002FstarVLA.git 克隆下来，在那个仓库下跑，本仓库仅供参考示意，提供相对原项目的增量代码）\n\n## 部署推理\n\n在 `starVLA\u002F` 下先起 policy server,启动前更换run_policy_server.sh的模型路径，换成自己训练的模型：\n\n```bash\nbash examples\u002FSonicLatent\u002Feval_files\u002Frun_policy_server.sh\n```\n\n再起在线推理：\n\n```bash\nPYTHONPATH=$PWD python examples\u002FSonicLatent\u002Feval_files\u002Frun_starvla_inference.py \\\n  --ckpt-path \u002Fplayground\u002FCheckpoints\u002Fsonic_latent_scratch_frozen_vlm\u002Fcheckpoints\u002F\u003Cckpt> \\\n  --host 127.0.0.1 \\\n  --port 10093 \\\n  --prompt \"pick up the cylinder and throw it into the trash bin\"\n  --rate 1.0\n```\n\n然后在 `wbc\u002F` 下依次启动\n\n```bash\npython gear_sonic\u002Fscripts\u002Frun_sim_loop.py --enable-image-publish --enable-offscreen --camera-port 5555\npython gear_sonic\u002Fscripts\u002Frun_camera_viewer.py --camera-host localhost --camera-port 5555\nbash deploy.sh --input-type zmq_manager sim\npython gear_sonic\u002Fscripts\u002Fsend_keyboard_cmd.py k \n```\n\nsend_keyboard_cmd作用与运行时机：\n\n- `k`: 先启动 deploy.sh 完毕后，机器人完全进入init状态，发送键k可让机器人进入CONTROL模式，机器人会在空中挣扎（没有挣扎的话重启deploy.sh），然后在MuJoCo界面按9可将其放下\n- `i`: 机器人放下后，发送键i可让机器人张开双手，准备执行任务（没有张开手的话再发送一次i）\n- `p`: 机器人准备好之后发送键p启动\u002F暂停 VLA policy \n\n比如要发送键`k`，直接在独立终端运行：\n\n```bash\npython gear_sonic\u002Fscripts\u002Fsend_keyboard_cmd.py k \n```\n\nsend_keyboard_cmd这个脚本请在gear_sonic_sim环境里运行。\n\n## 目录说明\n\n- `starVLA\u002Fexamples\u002FSonicLatent\u002F`: G1 VLA 训练和部署\n- `wbc\u002Fgear_sonic\u002Fscripts\u002F`: 采集、推理、仿真入口\n- `wbc\u002Fgear_sonic_deploy\u002F`: G1 部署代码\n","SonicStar 是一个针对 Unitree G1 机器人的开源视觉-语言-动作（VLA）堆栈，旨在支持遥操作数据收集、SonicLatent 训练、仿真以及实时全身策略部署。项目核心功能包括通过视觉和语言输入指导机器人执行复杂任务的能力，并提供了从数据采集到模型训练再到实际部署的完整流程。采用 Python 编写，分为数据采集与处理、模型训练及推理部署两大部分。适合于研究机构或开发者探索机器人在复杂环境下的自主导航与操作能力提升，尤其是在需要结合视觉感知与自然语言理解来实现精细动作控制的应用场景中。","2026-06-11 04:07:57","CREATED_QUERY"]