Overview
Learn one policy, not one controller
Participants develop a single control policy that operates the HVAC systems of commercial buildings across a range of layouts and weather conditions. The policy decides when to run each air handler and the hot-water pump, sets their temperature, pressure, and recirculation setpoints, and divides the air between zones, to jointly minimize energy cost and carbon emissions while maintaining occupant comfort.
The objective is deliberately not to tune a controller for one building. It is to learn one policy that generalizes across arbitrary floor plans, HVAC topologies, and weather. The same policy is deployed, unchanged, on buildings it has never encountered, differing in number of floors, geometry, room layout, zoning, equipment count, and climate. The policy must infer an effective strategy from the building description and live sensor stream it is given at runtime.
This portability across the entire distribution of buildings, rather than competence on a single instance, is the defining requirement of a building foundation model and the central scientific challenge of the competition.
The simulator, the training scenarios, the baseline controllers, the building generator, and a local scorer are in the starter kit on GitHub.
At each control step, a five-minute interval over a 30-day episode of 8,640 steps in which every step is scored, the policy observes the building state and must simultaneously issue:
- Discrete actions: an on/off command for each Air Handler Unit (AHU) and for the central Hot Water System (HWS), determining which equipment is actively conditioning the building.
- Continuous setpoints: for each AHU, the supply-air temperature, duct static pressure, and recirculation fraction; for the HWS, the boiler supply-water temperature and pump differential pressure; and for each zone's VAV box, the damper position and reheat valve position.
Because the number of AHUs and zones differs from building to building, both the observation and action vectors change shape per building, and there is no fixed input or output mapping the policy can memorize. This hybrid, variable-dimensional action space, combined with the requirement to generalize to unseen buildings, makes the problem substantially harder than standard fixed-dimension RL benchmarks.
Building components
A coupled dynamical system
Changing one AHU's setpoint affects multiple zones; occupancy in one zone shifts the thermal load across the building.
Air Handler Units
Central systems that cool and mix air and distribute it via ducts. One AHU serves at most 10 zones; the training buildings have 3 to 9 AHUs.
VAV Boxes
Terminal units in each zone. The policy sets each box's damper (the zone's share of air) and reheat valve (hot water to warm the air). They are the only per-zone controls.
Hot Water System
A central boiler and pump supplying heated water to AHU coils and zone terminals. The policy sets boiler setpoint and pump pressure to regulate heating capacity.
Zones
Individual rooms with independent temperature sensors and occupancy schedules. The training buildings have 28 to 85 zones, depending on size and layout.
Buildings differ not only in equipment counts but in how equipment maps to physical space. Each AHU serves a distinct group of rooms, and this mapping is unique to every building.
building_train_1: three floors, 39 zones, 4 AHUs (serving 10, 10, 9, and 10 zones). Both the number of AHUs and the zones each serves vary from building to building.
building_train_8 (two floors, 3 AHUs, 28 zones), building_train_1 (three floors, 4 AHUs, 39 zones), and building_train_2 (two floors, 9 AHUs, 85 zones). Footprint, floor count, zone count, AHU grouping, and envelope construction all vary.Observation & action spaces
Building-specific by design
Both spaces are determined by the floor plan, chiefly the number of zones and AHUs, and are not fixed across the competition.
State (observation)
Each observation's length scales with the building (it grows with zones and air handlers). It encodes:
- Per zone: air temperature, current heating and cooling setpoints, damper position, reheat valve position, airflow setpoint.
- Per AHU: supply-air temperature, static pressure, and airflow (sensed and setpoint), recirculation, a cooling-request count, and fan run status.
- Hot-water system: supply-water temperature (sensed and setpoint).
- Occupancy & schedule: building-level occupancy count, plus comfort-mode flags for now and imminently.
- Weather: outside-air temperature.
- Time: time of day and day of week (cyclically encoded).
Optionally, the policy may also consume the static floor plan and the AHU-to-zone mapping, supplied through the building metadata at runtime.
Action space
The action is a hybrid, variable-dimensional dictionary. For a building with K AHUs and V zones (one VAV box each):
- the discrete component has shape
(K+1), one on/off bit per AHU plus the Hot Water System; - the continuous component has shape
(3K+2+2V): three setpoints per AHU (supply-air temperature, static pressure, recirculation), two for the HWS (supply-water temperature, differential pressure), and two per VAV box (damper, reheat valve), each normalized to \([-1, 1]\).
An AHU sends one supply-air temperature to all its zones, so the dampers and reheat valves are the only per-zone decisions. The simulator enforces ventilation floors: a running AHU keeps a minimum static pressure, recirculation is capped at 70%, and every damper stays at least 10% open.
These shapes vary widely because the generator samples a broad configuration space: 2 to 4 floors; a 400 to 700 ft × 300 to 500 ft footprint; composite layouts; 30 to 60 ft rooms; and 1 to 2 occupants per zone. Across the 6 training buildings, K runs from 3 to 9 and V from 28 to 85, so the discrete space spans 4 to 10 bits and the continuous space 67 to 199 values.
Training building building_train_1 (Figure 2), with 4 AHUs and 39 zones, has a 5-bit discrete action and a 92-value continuous action. Read the index layout from the building metadata rather than assuming it.
Reward function
A regret-based, three-way tradeoff
At each step the policy receives a weighted sum of three competing quantities: comfort, energy cost, and carbon emission.
The comfort term \(\delta_t\) equals 0 when occupied zones are comfortable and falls as zones leave the comfort band: linearly up to 5 K outside the band, then quadratically, with no lower limit. So abandoning comfort is never free. Energy cost \(c_t\) and carbon \(e_t\) enter as costs. The organizer-fixed weights are identical for every submission and sum to one:
This makes comfort dominant while still charging materially for energy and carbon, and it creates the central tradeoff. Conditioning aggressively keeps comfort near its best value but drives energy and carbon high; switching equipment off eliminates energy and carbon but drives the comfort penalty up whenever zones are occupied. Because comfort is scored only in occupied zones, the reward favors forward-looking, occupancy-aware control: precondition spaces before people arrive, then relax toward an energy-saving band once they leave.
