The work, published February 11, 2026 in IEEE Transactions on Robotics, introduces a system called "rulebooks" that gives autonomous systems a principled, transparent method for resolving conflicts between goals such as safety, legal compliance, efficiency and passenger comfort. The research team includes Tichakorn Wongpiromsarn, associate professor of computer science at Iowa State; Konstantin Slutsky, assistant professor of mathematics at Iowa State; and Emilio Frazzoli, professor of dynamic systems and control at ETH Zurich.
Current autonomous systems typically handle competing objectives through a single mathematical cost function that assigns a numerical weight to each goal and selects whichever action scores highest across all of them. Wongpiromsarn said this approach breaks down in situations where certain priorities should be absolute rather than tradeable. If an efficiency weighting is set too high, a robot may drive aggressively; adjusting that weight does not resolve the deeper structural problem, which is that safety is being treated as just another factor to offset rather than a hard boundary.
A separate design approach divides goals into "hard" and "soft" constraints, with hard constraints taking priority regardless of cost. But that method also fails in genuine dilemmas. If a pedestrian steps in front of a self-driving car, the vehicle may have no option that satisfies the hard constraint of preventing harm -- it can either brake and risk hitting the person or swerve and risk a collision with oncoming traffic. A hard-versus-soft system has no mechanism for choosing between two options that both violate the top constraint; it can only declare the situation unsolvable.
The rulebooks framework avoids both problems by replacing weighted blending with explicit ranked ordering. Each rule represents a specific goal, and the system defines which rules take precedence, which are treated as equivalent, and which cannot be directly compared. When a conflict arises, the robot identifies which rules can still be satisfied and selects the action that performs best under the highest-ranked achievable constraints. Violations become comparable and the least harmful option becomes identifiable.
"This approach lets robots behave more like people," Wongpiromsarn said. "People typically follow the most important rules first and only consider lower-priority goals once the critical ones are met or proven impossible."
The framework also supports layered or incremental specification of priorities. Base-level rules can be set by regulators or law -- for example, a legal mandate that self-driving cars must avoid harming humans or property -- and manufacturers can then add further ranked goals, such as lane discipline or proximity to curbs, as long as those additions remain subordinate to the legal baseline. Slutsky said this structure allows regulatory compliance without over-restricting commercial design choices. "Everyone follows the same core rules, but companies still have the freedom to innovate and design their own behavior," he said.
A further benefit of the framework is transparency. Because goals are ranked rather than blended into a single opaque score, engineers, regulators and courts can examine a robot's reasoning after the fact and determine whether it acted in accordance with stated priorities. Wongpiromsarn said this audit capability is particularly important for post-incident analysis following crashes, near-misses, or regulatory reviews.
The researchers also demonstrated that rulebooks function as a common language capable of expressing multiple existing robot-control approaches -- logical rules, optimization objectives, and constraint-based methods -- within a single unified structure. In testing, their algorithms efficiently generated plans that respected complex priority structures and outperformed standard planning methods in scenarios where those methods broke down.
The team noted that the implications extend well beyond robotics. As AI systems take on more decision-making authority in transportation, health care, public safety and related domains, the ability to justify choices in terms of clearly ordered human values becomes more important. "The rulebooks concept offers a way to encode societal values, legal norms and organizational policies directly into machine decision-making," Wongpiromsarn said. "When machines make hard choices, they do so according to priorities humans can understand and even hold them accountable for."
Research Report: Formal Specification and Control Synthesis of Autonomous Robots Using Rulebooks
Related Links
Iowa State University
All about the robots on Earth and beyond!
| Subscribe Free To Our Daily Newsletters |
| Subscribe Free To Our Daily Newsletters |
