← Back to Videos

Blind Tight-Clearance Assembly from Proprioception Alone, Zero-Shot to Real Robots

Loading video

No cameras, no tactile or force-torque sensing, no priors on object geometry or pose — only joint position readings. Sim-trained blind policies insert pegs, mesh gears and thread nuts; the peg policy transfers zero-shot, inserting unseen pegs in 73% of real trials.

Manipulation Sim-to-Real Proprioception Peg-in-Hole Reinforcement Learning MIT

Category: arm

Author: @younghyo_park

Date: 2026-10-19T00:00:00

Duration: 35.6s

Source: https://x.com/younghyo_park/status/2112182379091501136

Reference: https://arxiv.org/abs/2610.23849

ACCRUE: Continual Off-Policy RL for Robot Foundation Models

ACCRUE: Continual Off-Policy RL for Robot Foundation Models

@cykim1006 · 102 views · 2026-10-14

Reinforcement Learning · Continual learning · Robot foundation models

AlphaNIST: Chaining Ten Robotic Assemblies from Sparse Rewards

AlphaNIST: Chaining Ten Robotic Assemblies from Sparse Rewards

@mateoguaman · 160 views · 2026-10-13

AlphaNIST · Reinforcement Learning · Robotic Assembly

AlphaNIST: Binary Reward Plus Control Cost Hits 98.2% Sim Success

AlphaNIST: Binary Reward Plus Control Cost Hits 98.2% Sim Success

@octi_zhang · 140 views · 2026-10-13

Robotic Arm · Manipulation · Reinforcement Learning

AlphaNIST: One Policy From Scratch Solves All 20 NIST Assembly Tasks

AlphaNIST: One Policy From Scratch Solves All 20 NIST Assembly Tasks

@patrickhyin · 110 views · 2026-10-13

Robotic Arm · Manipulation · Reinforcement Learning

Assembling Two Parts in One Hand

Assembling Two Parts in One Hand

@heetezition · 522 views · 2026-09-11

In-Hand Manipulation · Dexterous Assembly · Reinforcement Learning

Play2Perfect: Play Pretraining for Precise Assembly

Play2Perfect: Play Pretraining for Precise Assembly

@tylerlum23 · 710 views · 2026-07-01

Dexterous Manipulation · Robot Assembly · Reinforcement Learning

As an Amazon Associate, we earn from qualifying purchases.