EmbodiedIntentBench
A benchmark for evaluating runtime adaptivity of mobile robots. It quantifies task intent as structured contracts and causally couples perception decisions with physical execution and resource consumption, using task completion rather than perception accuracy as the evaluation criterion.
Timeline (last 90 days)
EmbodiedIntentBench is a robot evaluation benchmark that quantifies task intent into structured contracts (covering coverage, temporal evidence, reporting requirements, urgency, and resource constraints) and causally couples perception decisions with physical execution and resource consumption.
EmbodiedIntentBench binds simulated unmanned aerial vehicle (UAV) workloads to a causal, resource-accountable execution process, where computational decisions directly affect task time, resource consumption, and valid evidence.
In the multi-stage Depot study of EmbodiedIntentBench, a configuration that achieved 87.5% full-depot target recall on visible target observations completed only 1 of 16 tasks, whereas a configuration with lower recall completed 14.
In the ablation experiments of EmbodiedIntentBench, ignoring the observation loss caused by the execution process leads to an overestimation of the overall task success rate, and the ranking of the two runtime policies is reversed.
EmbodiedIntentBench uses CPU and GPU configuration replay on the Jetson AGX Orin, and the results show that the relative success rates of local and remote policies vary with the hardware configuration.
In some Jetson GPU configuration replays, EmbodiedIntentBench's intent-driven adaptive policy reduces modeled task energy consumption by up to 28% while keeping the task success rate unchanged.
All Facts (6)
EmbodiedIntentBench is a robot evaluation benchmark that quantifies task intent into structured contracts (covering coverage, temporal evidence, reporting requirements, urgency, and resource constraints) and causally couples perception decisions with physical execution and resource consumption.
50%UnverifiedEmbodiedIntentBench binds simulated unmanned aerial vehicle (UAV) workloads to a causal, resource-accountable execution process, where computational decisions directly affect task time, resource consumption, and valid evidence.
50%UnverifiedIn the multi-stage Depot study of EmbodiedIntentBench, a configuration that achieved 87.5% full-depot target recall on visible target observations completed only 1 of 16 tasks, whereas a configuration with lower recall completed 14.
50%UnverifiedIn the ablation experiments of EmbodiedIntentBench, ignoring the observation loss caused by the execution process leads to an overestimation of the overall task success rate, and the ranking of the two runtime policies is reversed.
50%UnverifiedEmbodiedIntentBench uses CPU and GPU configuration replay on the Jetson AGX Orin, and the results show that the relative success rates of local and remote policies vary with the hardware configuration.
50%UnverifiedIn some Jetson GPU configuration replays, EmbodiedIntentBench's intent-driven adaptive policy reduces modeled task energy consumption by up to 28% while keeping the task success rate unchanged.
50%