An Empirical Study of Reinforcement Learning for Keypoint Positioning of Deformable Linear Objects
An Empirical Study of Reinforcement Learning for Keypoint Positioning of Deformable Linear Objects
This research investigates reinforcement learning approaches for planar manipulation of deformable linear objects (DLOs) with one end grasped. The authors compare three Proximal Policy Optimization training stacks, finding that a GPU-parallel MuJoCo Playground setup offers faster convergence and consistent performance compared to CPU-based alternatives. By utilizing relative keypoint observations and an asymmetric actor-critic formulation, the study achieves simulation success rates of 96.4% for single-point tracking, 98.0% for disturbance recovery, and 88.8% for two-point tracking. Furthermore, policies trained with observation noise, delay, domain randomization, and disturbances successfully transfer to physical trials without fine-tuning, achieving success in 9 out of 10, 10 out of 10, and 8 out of 10 trials respectively using markerless RGB-D keypoint estimation.
The study empirically compares Proximal Policy Optimization training stacks for deformable linear object manipulation, demonstrating that GPU-parallel simulation with specific observation designs enables high success rates in simulation and successful zero-shot transfer to physical trials.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
Reported success rate
96.4%
Unique elapsed hours
View original evidence
prove learning efficiency. Policies trained with observation noise, delay, domain randomization, and disturbances achieve simulation success rates of 96.4% for single-point tracking, 98.0% for disturbance recovery, and 88.8% for two-point tracking. Using markerless RGB-D keypoint estimation, the policies transfer without real-world fine-tuning, succeeding in 9/10, 10/10, and 8/10 physical trials, respectively.
Open source S4
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
using on training efficiency, observation design, robustness, and zero-shot sim-to-real transfer. We compare three Proximal Policy Optimization training stacks and find that the GPU-parallel MuJoCo Playground stack enables faster convergence and more consistent performance than CPU-based alternatives under the same wall-clock budget. Relative keypoint observations and an asymmetric actor-critic formulation further im
Open source S3
prove learning efficiency. Policies trained with observation noise, delay, domain randomization, and disturbances achieve simulation success rates of 96.4% for single-point tracking, 98.0% for disturbance recovery, and 88.8% for two-point tracking. Using markerless RGB-D keypoint estimation, the policies transfer without real-world fine-tuning, succeeding in 9/10, 10/10, and 8/10 physical trials, respectively.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org