Multi-Level Hand Benchmarks

When building robot hands, the goal is to create a system capable of reliable dexterous manipulation across a broad range of tasks. Due to the system's complexity, it is difficult to understand how design decisions will affect a hand's dexterity. In other words, there is an attribution problem. To address this, we benchmark robot hand performance at four levels of varying complexity: application, system, hand, and component, with the goal of connecting component- and hand-level design decisions to system- and application-level performance. On this page, benchmarks at each of the four levels are listed along with implementation details.

For Hand and Component Level Benchmarks, the testbeds that are used to evaluate these benchmarks are listed in the Component Testbed tab.