Steering Law¶
An empirically calibrated human-movement law relating continuous bounded-path traversal time to the integral of reciprocal local corridor width.
Core Idea¶
Accot and Zhai's steering law models human movement through a continuously bounded path. For path \(C\) of local width \(W(s)>0\), its difficulty index is \(ID_C=\int_C ds/W(s)\), and fitted completion time is \(T=a+bID_C\). For a straight constant-width tunnel the index reduces to length over width, \(A/W\). The original stylus studies tested straight, narrowing and spiral tunnels. The law is an empirical HCI model, not an exact rule for all human motion, cars or robots.[^ref-fd50e26ae03a]
Scope of Application¶
Use the model when a human-controlled pointer must remain within a corridor all along its route. Accot and Zhai tested stylus drawing and proposed menu navigation as another path-steering application. They recorded boundary-crossing errors as well as successful-trial time. The coefficients \(a,b\) require calibration for the device and task; geometry alone does not set a universal duration.[^ref-fd50e26ae03a]
A single aimed reach to an endpoint belongs to the neighboring Fitts-law task family, whose difficulty index is logarithmic. The road-driving remark in the original paper is motivation, not validation of a vehicle-safety law. The live Fitts node is therefore related, not an asserted strict parent; this workspace entry remains unparented pending review.
Clarity¶
The integral preserves the cost of narrow sections that an average corridor width might hide. Equal path lengths do not imply equal predicted difficulty if their width profiles differ. At the same time, equal indices do not guarantee equal observed time or error across devices, users, curvature and feedback conditions. A model fit must be tested rather than assumed.[^ref-fd50e26ae03a]
Manages Complexity¶
One index describes straight, narrowing and curved tunnels, reducing to \(A/W\) only for constant width. It turns a complex route into additive local difficulty contributions, making path designs comparable without inventing a new score for each shape. The simplification does not eliminate the need to report errors and the conditions under which the time relation was measured.[^ref-fd50e26ae03a]
Abstract Reasoning¶
First establish that the task imposes continuous along-path boundaries. Then specify the route, arc-length coordinate and normal width \(W(s)\); compute \(ID_C\); fit \(a,b\) from comparable human trials; and check whether observed time and boundary errors support the model. Do not substitute the authors' approximate local speed–width relationship for an exact universal movement law.[^ref-fd50e26ae03a]
Knowledge Transfer¶
In the original stylus experiments, the tunnel is drawn on screen and the hand moves a stylus through it. In the paper's hierarchical-menu model, the pointer moves through linked vertical and horizontal corridors. The roles of path, local width and integrated index transfer, while the empirical coefficients and additional decision time do not automatically transfer. Other vehicles or autonomous controllers require new evidence rather than analogy.[ref-fd50e26ae03a][ref-d68f8afe1744]
[^ref-fd50e26ae03a]: Johnny Accot and Shumin Zhai, “Beyond Fitts' Law: Models for Trajectory-Based HCI Tasks,” CHI 1997, pp. 295–302, Experiments 2–4 and Design Implications. https://chi1997.acm.org/proceedings/paper/ja.html [^ref-d68f8afe1744]: Johnny Accot and Shumin Zhai, “Performance Evaluation of Input Devices in Trajectory-Based Tasks,” CHI 1999, publisher title/abstract scope. https://dl.acm.org/doi/10.1145/302979.303133
Neighborhood in Abstraction Space¶
Steering Law sits in a sparse region of the domain-specific corpus (60th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Usability & Interaction Design Failures (13 abstractions)
Nearest neighbors
- Fitts's Law — 0.86
- Pursuit Curve — 0.86
- Orientation Loss — 0.85
- Handoff Loss — 0.84
- Cognitive Walkthrough — 0.84
Computed from structural-signature embeddings · 2026-10-08