Xiaomi Robotics Team Sweeps Dual International Championships, Outperforming Runners-Up by Wide Margins
Xiaomi's robotics team has achieved top rankings at two major international competitions – CVPR 2026 and ICRA 2026 – with models that far surpassed second-place entries. The team's WAM model 'my16' recorded a 40.89% success rate, the only model above 40%, while the ICRA benchmark saw a 94% overall success rate, leading by 10 percentage points.
Xiaomi’s robotics team has secured first place in two prestigious international competitions this year, with results that significantly outpaced the runners-up, the company’s founder and CEO Lei Jun announced on Thursday.
CVPR 2026 Workshops: RoboChallenge
At the CVPR 2026 Workshops – GigaBrain Challenge RoboChallenge Track, Xiaomi’s anonymously submitted model “my16” achieved a 40.89% success rate across 30 extremely difficult real-world tasks. This was the only model in the competition to break the 40% threshold, ranking first overall with a clear lead over all other entries.
The tasks included bimanual dexterous manipulation, covering flexible objects, tool causal reasoning, and cross-platform robustness. Each task required 10 consecutive uninterrupted tests, with participants submitting a unified multi-task model to test generalization capability.
The winning “my16” model is a World Action Model (WAM) developed in-house, featuring a “brain+ cerebellum + long-short term memory” architecture. It integrates three standout capabilities into one unified system:
- S1/S2 dual-system architecture for cognitive depth,
- Long-short term memory for long-horizon stability,
- Cross-embodiment pre-training for controller-level execution precision.
This enables the model to combine the reasoning depth of large models with the execution accuracy of traditional controllers, while maintaining stability over extended operation sequences.
ICRA 2026 WBC: Supermarket Beverage Retrieval
In the ICRA 2026 Whole Body Control (WBC) competition, the challenge simulated a supermarket scenario. Robots had to recognize and pick specified beverages from shelves among 16 large categories and 20 subcategories of drinks, then place them into a shopping cart.
Xiaomi’s solution scored 99.2 points out of 100, near-perfect, with an overall success rate of 94% – the only entry exceeding 90% and a full 10 percentage points ahead of the second-place entry.
- Simple tasks: 100% success rate
- Complex tasks: 90% success rate
Lei Jun emphasized that the team’s direction is to bring robots into the physical world to perform real practical work.
For more technical details, Xiaomi Technology’s official Weibo account published a comprehensive breakdown of the models and competition results.
Source: IT Home, June 5, 2026