Industrial objectives become useful only when a system can translate them into appropriate decisions and measurable responses. This program studies planning, optimization, and control execution under defined constraints. It connects operational intent with permitted commands, physical feedback, and the conditions that require a stop, fallback, or human decision.
A recommendation, an issued command, and an achieved outcome are different events. A decision method may propose a lower energy setting, but equipment limits, operating state, permissions, and existing control logic determine whether that proposal can be applied. Observed behavior is then needed to evaluate what actually happened.
Consider a proposed cooling-system study that aims to reduce energy use while maintaining a defined temperature range. Candidate setpoints would be evaluated against equipment limits and observation freshness before any authorized change. Independent measurements would compare the resulting temperature and energy use with a baseline. This is an illustrative research question, not a completed experiment or a claim of energy savings.
Investigate software agents that interpret a bounded operational task, assess available evidence, and coordinate permitted tools or actions. Research questions include task decomposition, command preconditions, approval boundaries, and recovery from incomplete execution. An agent should distinguish a proposed action from an authorized command and a confirmed physical outcome, with records that allow those transitions to be inspected.
Study planning and control methods that account for equipment limits, process requirements, and uncertainty. Candidate approaches may include optimization and model predictive control, which uses a model to evaluate future behavior before selecting an input. Model assumptions and feasible operating regions need explicit treatment; satisfying constraints in a model does not guarantee that real equipment will remain within them.
Examine coordinated operation of heating, ventilation, air conditioning, and related energy equipment. Questions include changing loads, temperature requirements, actuator limits, and integration with existing controllers. NIST research provides examples of developing and evaluating AI control algorithms alongside physical HVAC systems and test infrastructure. Comparable research questions require an explicit baseline and documented operating conditions.
A proposed evaluation would begin with a defined objective, baseline controller, permitted actions, reference measurements, and stop conditions. Replay and simulation could examine candidate behavior before physical testing. A supervised equipment test, where justified, should compare commanded inputs with measured responses, record rejected or incomplete commands, and examine changed loads, stale observations, and interrupted communication.
Measures may include objective achievement, tracking error, constraint violations, decision latency, rejected commands, recovery time, and intervention frequency. Energy studies should report comparable loads and conditions alongside consumption. Operator usefulness and economic value require separate evidence. Numerical thresholds need technical approval, and neither a successful simulation nor a command acknowledgement establishes a successful physical outcome.
This overview describes the program’s research agenda. It does not report completed trials, demonstrated autonomous operation, supervised pilots, or deployment. Approved project records should identify dated status, methods, setup, findings, limitations, and next steps. No performance figures, guaranteed behavior, or certification claims are asserted here.
Planning, optimization, and control execution belong to this program. Broader assurance methods, authorization security, and simulation infrastructure have separate owners. Existing protective functions and operator responsibilities must be addressed explicitly in the design of any proposed test.
Visit the specialist control website for detailed program research. To discuss a challenge, describe the equipment, operational objective, existing controller, observations, permitted actions, constraints, and outcome you need to evaluate. Begin with a nonconfidential summary.
All Research Programs