ResearchResearch paperRobotics & Embodied AI · Evaluation & Benchmarks2 sources · Oct 18, 2026

BiDex-100: A Comprehensive Benchmark for Bimanual Dexterous Manipulation Across 100+ Simulation Tasks

In this paper, we present BiDex-100, a simulation benchmark for bimanual dexterous manipulation that combines broad task coverage, multi-dimensional evaluation, and high-DoF hand embodiments.

Key points

  • Bimanual manipulation with dexterous hands has attracted growing attention in research on robotic action models.
  • However, few studies have developed benchmarks specifically tailored for dexterous manipulation, leaving the dexterous manipulation capabilities of existing models insufficiently characterized.
  • In addition, the benchmark supports two embodiments and provides 8,080 success-validated teleoperated demonstrations, totaling 6.25 million frames.
  • We evaluate seven recent Vision-Language-Action models under a unified protocol, establishing baselines for capability-specific performance, robustness, and data efficiency.

Sources (2)

Extractive summary: sentences quoted from the sources.

Before this

  1. Oct 16, 2026NestedVLA: Learning to Consolidate and Generate Skills for Vision-Language-Action Models
  2. Oct 16, 2026MambaAct: Efficient Mamba Action Experts for Flow-Matching Vision-Language-Action Models
  3. Oct 16, 2026Persistent Planar Memory for Video World Models
  4. Oct 15, 2026When Plans Change, Remember: Replanning-Guided Visual Memory for Vision-Language-Action Models
  5. Oct 15, 2026huggingface/transformers v5.20.0: Release v5.20.0
  6. Oct 15, 2026REKI: Registering Knowledge in VLAs

Related