Xi Wang

Current manipulation policies excel at basic tasks but falter in complex skill execution and generalization, revealing critical gaps in their real-world applicability.

MV-Actor achieves a remarkable 87.8% success rate in bimanual manipulation by effectively sharing semantic perception across multiple camera views, outpacing traditional methods.

Massive multimodal models like Qwen and CLIP excel at information retrieval, but their sheer size makes them impractical – this workshop tackles the efficiency gap.

Trace generation, not proving, is now the major bottleneck in zkVMs, but a new accelerator design blows away CPU performance by over 1800x.

Automating UVM testbench generation with LLMs slashes verification time from days to hours, achieving near-complete code coverage.