Ant's Robbyant Open-Sources LingBot-Touch Tactile Foundation Model

Ant Group's Robbyant team (Zhipingfang) has released the open-source tactile foundation model LingBot-Touch for dexterous manipulation. It's on Hugging Face under Apache 2.0. Built on a JEPA architecture with 5.46 million parameters, the model was pretrained on nearly 5,000 hours of real tactile data collected from gloved human hands and robot fingertip sensors.

Confirmed

• Available in Tiny and Small sizes, open-sourced on Hugging Face under Apache 2.0
• Unified heterogeneous tactile modeling enables transfer across sensors, dexterous hands, tasks, and action models
• Per @AlternativeSet4042, real-robot deployment paired with pi0.5 raised the task score from 48 to 70
• The model is a 5.46-million-parameter tactile encoder

Why it matters

• Internet data taught models to see and speak, but not what it feels like to hold a cup; LingBot-Touch targets exactly this gap
• @AlternativeSet4042 notes that two prior preprints applied GPT-6 Astra tactile images to manipulation tasks, but were validated only in simulation, with one taking a median 14 seconds per call; LingBot-Touch's small footprint and real-robot validation stand in contrast, offering a practically deployable open-source baseline for tactile manipulation.

2026-11-10 ~ 2026-11-11 · 30 related posts

• [source] 5.5M-param tactile encoder hits real hands, lifts pi0.5 scores from 48 to ~70 — Alternative_Set4042 · 2026-11-10
• Ant Group's Robbyant open-sources LingBot Touch, a JEPA-based tactile foundation model — AdinaYakup · 2026-11-10
• Open-source tactile foundation model LingBot-Touch trained on ~5,000 hours of real data — Shruti_0810 · 2026-11-10
• Ant's LingBot-Touch: open-source tactile foundation model boosts dexterous manipulation across 26 tasks — 新智元 · 2026-11-10
• Robbyant's LingBot-Touch: a tactile foundation model trained on 5,000 hours of real data — 机器之心 · 2026-11-10
• Robbyant releases LingBot-Touch, a tactile model giving robot hands human-like touch sensing — CyberRobooo · 2026-11-10
• LingBot-Touch unifies three tactile sensor types using 4,983 hours of real recordings — CyberRobooo · 2026-11-10
• How LingBot-Touch works: one shared Haptic Transformer unifies magnetic, piezoresistive and visuotactile signals — CyberRobooo · 2026-11-10
• LingBot-Touch lifts VLA success rate from 45.6% to 72.5%, pretrained on 4,983 hours of tactile data — CyberRobooo · 2026-11-10
• Ant Lingbo open-sources LingBot-Touch: tactile tokens lift robot zipper success from 15% to 59% — 赛博禅心 · 2026-11-10
• [source] LingBot-Touch tactile foundation model unifies heterogeneous sensors for dexterous manipulation — robbyant · 2026-11-10
• LingBot-Touch: a foundation model for heterogeneous tactile representation learning lands on HF — _akhaliq · 2026-11-10
• LingBot-Touch links: technical report, GitHub repo, and Hugging Face all public — aigclink · 2026-11-10
• Robbyant open-sources LingBot-Touch, a JEPA-based tactile foundation model for dexterous manipulation — pmttyji · 2026-11-10
• Ant's Robbyant trains LingBot-Touch: 2.5h of labels teach robot hands to sense force — heyshrutimishra · 2026-11-10
• Ant Group's Robbyant gets LingBot-Touch to read force it was never taught — heyshrutimishra · 2026-11-10
• Robbyant open-sources LingBot-Touch, giving robots fingertip touch sense with JEPA encoders — rohanpaul_ai · 2026-11-10
• Robbyant open-sources LingBot-Touch, a tactile foundation model trained on 4,983 hours of data — rohanpaul_ai · 2026-11-10
• Ant-backed open-source touch encoder lifts π0.5 robot success from 48% to 70% — kimmonismus · 2026-11-11
• Learned touch tokens add 14–25 points over raw signals across four policies — kimmonismus · 2026-11-11
• 5,000 hours of paid touch data: robotics' scarcest, unscrapeable training input — kimmonismus · 2026-11-11
• One touch encoder generalizes across sensors, cutting contact error to 0.16mm — kimmonismus · 2026-11-11

8 near-duplicate retellings: dair_ai · Div_pradeep · HeyAmit_ · aftahi_ai · alifcoder · aigclink · omarsar0 · kimmonismus