EEBoundsAndSafety: default bounds [-1,1] m silently no-op on small arms — require explicit bounds or warn loudly

[johnnynunez]

Context

EEBoundsAndSafety ships with default bounds of [-1, 1] m. For small arms (SO-101 reach ≈ 0.35 m) that is ~3× the robot's workspace — so the safety node never engages while giving the integrator the impression that a safety layer is active. Worst of both worlds: perceived safety, none delivered.

Evidence

From per-frame CSV instrumentation over a week of SO-101 sessions:

• 85% of frames commanded EE targets outside the physically reachable envelope; ee_z peaked at 0.91 m on a 0.35 m-reach arm.
• The stock node did not intervene once all week.
• Out-of-workspace targets are what drive the 5-DOF IK to oscillate between solution branches (measured ±100°/frame) — i.e., the un-engaged bounds check sits exactly upstream of the worst instability we observed.

Proposal

• Require explicit bounds (no default), or
• Fail loudly when defaults are used: a startup WARNING like EEBoundsAndSafety using default bounds [-1,1] m — these are almost certainly wrong for your robot; measure and set them would have saved us days.

Optionally: accept bounds from the robot description (URDF/USD reach) where available.

---

Part of the field-integration feedback series (SO-101 + reBot DevArm, June 29 – July 6). Priority suggestion: P1.