roto 2.0: The Robot Tactile Olympiad

2026-05-20

AI Summary

This work addresses the long-standing fragmentation in tactile reinforcement learning, which has predominantly focused on saturated pose-centric tasks and lacked standardized benchmarks. To bridge this gap, the study introduces the first purely tactile-driven “blind manipulation” benchmark that relies solely on proprioceptive and tactile inputs—requiring neither ground-truth state information nor policy distillation—to enable end-to-end control. The benchmark supports four robot morphologies with 16–24 degrees of freedom and leverages GPU-accelerated parallel simulation, offering an open-source environment alongside well-tuned baselines to substantially lower the barrier to entry. Empirical results demonstrate that agents trained within this framework can execute 13 rotations of a dexterous ball within 10 seconds, achieving a tenfold improvement in speed over the current state-of-the-art methods.

Abstract

Tactile-based reinforcement learning (RL) is currently hindered by fragmented research and a focus on over-saturated orientation tasks. We introduce v2 of the Robot Tactile Olympiad (\texttt{roto 2.0}), a GPU-parallelised benchmark designed to standardise tactile-based RL across four distinct robotic morphologies (16-DOF to 24-DOF). Unlike prior benchmarks, roto focuses on end-to-end "blind" manipulation, utilising only proprioception and tactile sensing without state information or distillation. We demonstrate a significant performance leap, with our blind agents achieving 13 Baoding ball rotations in 10 seconds, an order of magnitude faster than current state-of-the-art speeds. By open-sourcing our environments and robustly tuned baselines, we reduce the barrier to entry and enable researchers to prioritise fundamental algorithmic challenges over tedious RL tuning. Website: https://elle-miller.github.io/roto/

Problem

• tactile-based reinforcement learning
• Robot Tactile Olympiad
• blind manipulation
• benchmark standardisation
• robotic morphologies

Innovation

• tactile-based reinforcement learning
• blind manipulation
• GPU-parallelised benchmark
• robotic morphologies
• end-to-end control