Learning to Assemble the Soma Cube with Legal-Action Masked DQN and Safe ZYZ Regrasp on a Doosan M0609
Jaehong Oh, Seungjun Jung, Sawoong Kim · 2025
Read Learning to Assemble the Soma Cube with Legal-Action Masked DQN and Safe ZYZ Regrasp on a Doosan M0609 by Jaehong Oh, Seungjun Jung, Sawoong Kim (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-21272.txt.
Description
Jaehong Oh, Seungjun Jung, and Sawoong Kim teach a robot arm to build the Soma cube, a puzzle made of seven odd-shaped pieces. Their 2025 work uses a Doosan robot arm with a simple gripper. The hard parts are the huge number of moves, the risk of unsafe arm paths, and the need to learn a sound order for the pieces.
The team trains the arm with a learning method that hides all illegal moves, so it only tries moves the rules allow. A safe grip method lets the arm turn pieces without hitting risky arm poses. Training moves from two pieces to three and then to all seven. The arm solved the easy stage every time, and the full puzzle in about four of ten runs.
Catalog record
| CREATOR | Jaehong Oh, Seungjun Jung, Sawoong Kim |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 88 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-21272