SymmGrid: Super-Scaling Real-Robot Deep Reinforcement Policy Learning with Branched Symmetries

arXiv Video SymmGrid Code Robot Controller Code Docs Hugging Face Tutorials Got it Pub
Gabe Everett1*, Brice Gunter 1*, Ryan Vander Stelt2*, Cleiver Ruiz-Martinez1, Blake Hull3, Juan Rojas1
*Equal Contribution; 1Department of EECS, Lipscomb University; 2Department of CS, Lipscomb University; 3Department of ME, Lipscomb University;

Deep reinforcement policy learning directly in physical robots (on-robot learning) remains bottlenecked by slow wall-clock training times. We present SymmGrid, a trajectory-level augmentation framework inspired by parallelized symmetries that super-scales group transformations to significantly accelerate on-robot learning in both egocentric and exocentric visual setups. We model a Markov Decision Process (MDP) under a symmetry tree, in which state–action pairs have admissible parallelized invariant transformations that yield a geometric grid structure. The state is modeled with ego- or exocentric images and proprioception information. The latter require special treatment, in the form of homographies, to warp visual scenes in line with their corresponding spatial transformations. These parallelized transformations produce a large set of unique symmetric equivalences that populate the replay buffer with diverse and consistent experiences that speed up learning and improve performance. We present extensive training and evaluations performed directly on real robot manipulation contact tasks including peg-insertions, cable routing, and object relocations. Relative to SOTA, SymmGrid achieved wall-clock training convergence speedups of 1.37-2.17x, evaluation success rate improvements of 1.09x-1.27x, fastest training convergence times of 16.6, 10.9, and 79.3 minutes respectively. For trajectory wide assessments, we used normalized area under the curve (nAUC) ratios. SymmGrid achieved improvements of up to 2.59x. These results confirm that simple branch symmetries can have an outsized result due to super-scaling and bring us closer to sub-10 minute on-robot learning training in manipulation tasks suitable for arms and humanoids. The project page is available at https://symmgrid-robot.github.io/.

Uncut Training Process

Peg Insertion

Cable Routing

Object Relocation

Have you used SymmGrid successfully? Send us your videos to juan.rojas@lipscomb.edu

Zero-shot Robustness to Perturbations and Distractors

Peg Insertion

The agent successfully inserts the block after being blocked by a hand.

The agent successfully inserts the block after having its camera blocked.

Object Relocation

The agent repeatedly attempts to take the ball from a human hand before time runs out.

The agent and classifier both handle rotated plates during object relocation.

Emergent Real-World Grasp Behaviors

The robot learns multiple distinct grasp strategies directly on hardware. These emergent behaviors are difficult to reproduce in simulation because they depend on real contact dynamics, sensor noise, and robot-specific physical interactions.

BibTeX

          
            @misc{everett2026symmGrid,
                title={SymmGrid: Super-Scaling On-Robot Learning with Parallelized Symmetries and Egocentric–Exocentric Visual Perception.},
                author={Gabe Everett, Brice Gunter, Ryan Vander Stelt, Cleiver Ruiz-Martinez, Blake Hull and Juan Rojas },
                year={2026},
                eprint={2607.26985},
                archivePrefix={arXiv},
                primaryClass={cs.RO}
            }