urdf_validator v1.0 catches physics errors that syntax checkers miss

Every ROS 2 developer knows the pain: check_urdf passes your file, but Gazebo launches and the robot collapses. That's because check_urdf only validates syntax, not physics. A 2024 empirical study (ROBUST, Empirical Software Engineering) found that semantic errors like bad mass, inertia, or kinematic specs are among the most common real-world failure modes across open-source robot packages. Notlord69's urdf_validator fills that gap. The pip-installable tool (pip install urdf-validator) checks beyond schema: full-body center of mass, gravity torque per joint, motor effort margins (with PASS/WARN/FAIL summaries), support polygon and stability margin, plus Monte Carlo forward-kinematic reach envelope. It also supports payload-augmented statics (e.g., --payload-mass 5.0) and an optional MuJoCo cross-validation mode (--deep) for simulated-confidence sanity checks.

The tool's design is deliberately deterministic and AI-agent-callable. Rather than using heuristics that can fail silently, urdf_validator lets users explicitly declare robot type, contact links, or arm roots/tips, and warns if the heuristic disagrees. It clearly distinguishes between UNKNOWN (couldn't determine) and N/A (doesn't apply). For LLM-powered robotics workflows, the tool provides a structured Python API (api/task_runner.py) that returns PASS/FAIL/N/A/UNKNOWN per sub-check with a traceable geometric reason. No ML or LLM is used in the physics pipeline — the tool remains an auditable oracle. Current rough edges: humanoid foot-contact stability is always UNKNOWN without explicit --contact-links, mimic joints are treated as fixed, SDF is not supported, and --pose home falls back to zero. Full examples for TurtleBot3, PR2, ANYmal, Spot, and Franka Panda are in the README. MIT licensed.