Struct PowerManager 

pub struct PowerManager { /* private fields */ }

Power manager

Implementations

impl PowerManager

pub fn new() -> Self

Create new power manager

pub fn set_orchestration_state(&mut self, state: AmbientOrchestrationState)

Set the current orchestration state so power estimates track the real state machine in orchestrator.rs (ModelLifecycle).

pub fn set_active_model_count(&mut self, count: usize)

Set the number of models currently active on the managed device.

pub fn can_execute(&self, _device: &AmbientDevice) -> bool

Check if device can execute task

pub fn update_power_consumption( &mut self, device: &mut AmbientDevice, execution_time: Duration, )

Update power consumption after executing a task on a device.

Uses the power optimizer to record the power state transition and the power policy to determine the power budget.

pub fn power_policy(&self) -> &PowerPolicy

Get the current power policy.

pub fn set_power_policy(&mut self, policy: PowerPolicy)

Set the power policy.

pub fn power_optimizer_mut(&mut self) -> &mut PowerOptimizer

Get a mutable reference to the power optimizer for direct optimization.

pub fn battery_monitor_mut(&mut self) -> &mut BatteryMonitor

Get a mutable reference to the battery monitor for updates.

pub fn thermal_monitor_mut(&mut self) -> &mut ThermalMonitor

Get a mutable reference to the thermal monitor for updates.

pub fn get_battery_level(&self, _device_id: &str) -> f64

Get battery level

pub fn get_thermal_state(&self, device_id: &str) -> ThermalState

Get thermal state derived from the current estimated power draw.

Power-to-thermal mapping (mobile SoC heuristic):

• < 3W → Normal (cool)
• 3–7W → Warm
• > 7W → Critical

pub fn get_power_consumption(&self, device_id: &str) -> f64

Get power consumption in watts.

On platforms exposing a power API (e.g. RAPL on Intel, the Energy Meter on Android) this would query the hardware. On every other target we estimate consumption from the current orchestration state and the number of active models, which is what battery-aware ML scheduling and thermal management rely on:

State	Base power
Idle	~0.5 W
Active inference	~5.0 W + (active_models × 2.0 W)
Scrubbing	~3.0 W
Streaming	~4.0 W

pub fn estimate_battery_life_remaining( &self, current_battery_pct: f64, battery_capacity_wh: f64, ) -> f64

Estimate battery life remaining in hours.

hours = (battery_capacity_wh * current_battery_pct / 100.0) / power_consumption

Returns 0.0 if the estimated power consumption is zero (avoids division by zero) or if the battery percentage is non-positive.

pub fn should_throttle_inference(&self) -> bool

Decide whether inference should be throttled.

Returns true when the thermal state is Critical or when the estimated battery life (using the battery monitor’s current charge against a 15 Wh mobile battery as a reasonable default) drops below 1 hour.

pub fn get_power_metrics(&self) -> PowerMetrics

Aggregate the current power/thermal/battery snapshot.

estimated_battery_hours is Some when a non-zero battery capacity is known; here we use the battery monitor’s current level against a 15 Wh mobile battery default. Returns None when the device has no battery (e.g. mains-powered embedded host).