Fifteen teams took part in Robbyant's LingBot-VLA 2.0 challenge, adapting a vision-language-action model to real robots within 48 hours before a final in Shanghai.
A 48-hour format is a useful stress test for a specific claim: that foundation models for manipulation can be adapted to new hardware quickly. If adaptation takes months of data collection per robot, the economics do not work.
The finals reportedly exposed difficulties in data quality, perception under variable lighting, and execution precision - the same three failure modes that appear in most real deployment reports.
Hackathon formats flatter demos and punish reliability, so the results should be read as a measure of how fast teams can reach a working prototype rather than how close the field is to deployment.
The broader value is the benchmark itself: shared tasks on physical hardware give the community something better than vendor videos to compare against.
