Evaluation of Robo-Dopamine on RoboRewardBench

Evaluate the performance of the Robo-Dopamine checkpoints using the RoboRewardBench benchmark once the dataset and checkpoints are released, to assess their reward prediction accuracy relative to other vision-language reward models.

Background

Robo-Dopamine is a concurrent approach focused on process reward modeling for high-precision robotic manipulation. Comparing its checkpoints on a standardized benchmark would clarify its effectiveness relative to other models evaluated in this work.

At the time of writing, the Robo-Dopamine dataset and checkpoints were not publicly available, preventing direct benchmarking on RoboRewardBench and leaving a concrete evaluation gap.