Search papers, labs, and topics across Lattice.
DeepInsight II addresses the fragmentation in evaluating embodied AI systems by quantifying performance across navigation and manipulation benchmarks while maintaining a unified substrate. It reproduces results from released checkpoints and introduces MotionBench to standardize whole-body controller performance under a single workload, facilitating a direct comparison between simulation and real-robot trials. This empirical continuity allows for a more coherent understanding of the sim-to-real gap and enhances the diagnosis of physical execution issues through evidence-grounded repair actions.
Bridging the gap between simulation and real-world performance, DeepInsight II reveals how standardized evaluation can enhance the reliability of embodied AI systems.
Across a Physical AI stack, evaluation maturity is inversely aligned with deployment risk: foundation models enjoy mature, standardized harnesses, while the embodied layers on which deployment actually turns remain fragmented across benchmark-specific simulators, embodiments, and interfaces. The first DeepInsight report (v1) unified evaluation across this stack behind three abstractions---task, resource, and result---but its quantitative evidence centered on the foundation-model layer; navigation and manipulation (System 1) and whole-body control (System 0) remained simulation case studies, and physical execution was outside its empirical scope. DeepInsight II keeps that substrate fixed and quantifies the embodied half. First, it reproduces released-checkpoint references across two navigation and four manipulation benchmarks under their native protocols. Second, MotionBench places four released whole-body controllers under one workload and metric contract, then carries a qualified within-family cohort from parallel simulation to matched real-robot trials in which simulated and physical rollouts share a parent trace identity while retaining execution-domain-specific records, making the sim-to-real gap a native reduction rather than a reconciliation across toolchains. Third, a composed System 2--1--0 study extends trace localization into five evidence-grounded handoff labels, each mapped to a concrete repair action, with a measured repairability criterion and physical episodes testing the same attribution under hardware-observable state. The contribution is therefore not a new evaluation architecture, but empirical continuity from benchmark execution to matched robot evidence and repair-oriented diagnosis.
