MuJoCo Ackermann Steering

Problem

MuJoCo cars approximate Ackermann steering with a quartic joint-equality polynomial; picking the coefficients badly causes visible steering misalignment across the range.

What I built

• Derived the quartic polycoef two ways — Taylor expansion of the exact geometry and a least-squares fit over the steering range.
• Built an error table (SSE / RMSE / max error) and argued why RMSE is the right metric to optimize (sample-count-independent, whole-range, in degrees).
• Exposed wheelbase, track width, steering lock, and sample count as CLI parameters with a comparison plot.

Impact

• Produces a ready-to-paste MuJoCo `polycoef` vector for a given car geometry.
• The least-squares fit minimizes typical (RMSE) steering error across the full lock range.

Stack