Reinforcement Learning

Reward-driven robot training — simulation-heavy locomotion policies, real-world fine-tuning, and the sample-efficiency problem.

Coverage (5 articles)

SimTac & DexSkin: Biomorphic Tactile Arrays and Hybrid MPC for Sim-to-Real — November 2025 arXiv paper SimTac (Zhang et al., King’s College London/BIT) introduces a physics-based MPM simulator for vision-based biomorphic tactile sensors
Pi0.7 and Helix: VLA Transformers Advance Zero-Shot Dexterous Manipulation — Physical Intelligence's π₀.7 (April 2026) and Figure AI's Helix (February 2025) represent state-of-the-art Vision-Language-Action (VLA) models, achieving open-w
ThorArena Benchmark Exposes Force-Aware Gaps in Humanoid Controllers — ThorArena, a new benchmark from TU Munich, tests how humanoid robots handle physical contact by replaying real human forces in simulation. The results expose ma
NVIDIA Isaac GR00T 1.7 VLA Powers End-to-End Humanoid Policy Training — NVIDIA Isaac GR00T 1.7 VLA and full humanoid development platform launch July 2026, enabling neural feedback loops, zero-shot generalization, and torque mapping
Apptronik Robot Park Powers Apollo 3 Humanoid Data Factory — Apptronik's 90,000 sq ft Austin Robot Park, opened June 30 2026, deploys Apollo 2 fleets to generate real-world datasets for Gemini Robotics VLA models, acceler