PhysRSI-eval
PhysRSI-eval is a benchmark for evaluating the physical recursive self-improvement capability of vision-language models; it employs a five-axis metric system to measure the level of self-improvement of agents on physical manipulation tasks.
Timeline (last 90 days)
The paper proposes the PhysRSI-eval benchmark, employing a five-axis metric system: first-attempt performance, efficiency, success rate, budget consumption, and the ability to improve from failure.
The five-axis metrics can distinguish systems that have similar average scores on public benchmarks.
All Facts (2)
The paper proposes the PhysRSI-eval benchmark, employing a five-axis metric system: first-attempt performance, efficiency, success rate, budget consumption, and the ability to improve from failure.
50%UnverifiedThe five-axis metrics can distinguish systems that have similar average scores on public benchmarks.
50%