microduck_rl
microduck_rl 安全与运维：隔离仿真并保护硬件
治理资产、检查点、遥测、控制接口和事件恢复。
你将学会
- Map environment, simulator, and runner
- Run a reproducible simulation
- Identify sim-to-real gates
开始前需要
- Basic Git and command-line usage
- Comfort reading a project README
You can explain the project, run its documented first step, and decide what to verify before adopting it.
先看结论
- Simulation credentials and hardware controls are different trust zones.
- Watchdogs and stops must be independent.
- Telemetry and checkpoints need retention and incident plans.
Protect assets and data
训练代码默认不能访问硬件；隔离网络，限制执行器，设置独立看门狗和急停，并为异常动作演练停止、隔离和恢复。
Use the pinned source evidence, record versions and assumptions, and ask a qualified reviewer to validate the result before adopting it.
Constrain actions
Training code should not reach hardware by default. Separate simulation, export, and control networks; require allow-listed interfaces, actuator limits, watchdogs, and operator approval for trials.
Recover
Monitor divergence, crashes, runaway actions, disk growth, and stale checkpoints. Rehearse stop, revoke, quarantine, restore, and deletion with an independent emergency stop.
Use the pinned source evidence, record versions and assumptions, and ask a qualified reviewer to validate the result before adopting it.
如何选择
| 比较维度 | 方案 A | 方案 B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
实施步骤
- 1
Inventory assets, interfaces, credentials, and owners.
- 2
Isolate simulation and hardware networks.
- 3
Gate export/control with limits and approvals.
- 4
Monitor, stop, quarantine, and restore.
可复制示例
simulation worker -> bounded policy
export -> reviewed adapter + limits
control link -> watchdog + emergency stop
incident -> stop/revoke/quarantine/restore常见问题
Can a policy process own the emergency stop?
No. Use an independent, qualified stop mechanism controlled by the robot safety system/operator.
Should checkpoints be public?
Only after reviewing asset licenses, telemetry privacy, and potential misuse.
资料来源
- microduck_rl repository来源核查 2026-08-31