DSD

We compare DSD with standard action binning to test whether separating motion direction from magnitude improves real-robot task success without additional robotics pretraining. We reimplemented both representations on the same Florence-2 policy, starting from its public weights and following the paper’s parallel action decoding. Both arms train for 40,000 steps at batch 32 on one GPU, and the two arms differ only in the action representation.

• Longer chunks. We predict 30 steps instead of 20 because controls run at 30 Hz, giving a one-second action chunk.
• Absolute-command execution. Predicted relative increments are accumulated into absolute commands. The first increment is measured from the observed pose and has separate normalization statistics; later increments are differences between successive commands. Both arms use this construction.
• Wider normalization bounds. Both arms use the 0.1st/99.9th percentiles instead of the paper’s 1st/99th, for BIN coordinates and DSD magnitudes alike. With 1st/99th bounds, BIN alone clipped fast motions in our data, so even perfect tokens ended its one-second chunks about 15 to 25 mm off at the 95th percentile against about 1.5 mm for DSD; with the wider bounds both are near 1 mm.
• Exact-zero decoding. Motion bins containing zero decode to exactly zero instead of their bin centers, preventing idle drift when increments are accumulated. This replaces reliance on a controller dead zone and applies to both arms; grippers are unchanged.
• Direction-loss weighting. Our DSD weights the direction loss with a soft ramp, w = clip(s/τ, 0.1, 1), with τ = 0.3% of the full-scale range. The authors’ implementation uses a step: weight 0.1 below the threshold and 1.0 above it. In the authors’ LIBERO experiments, switching to the ramp alone reduced success by about 3 points, bringing DSD close to their binning baseline. The authors also recommend a lower threshold for our action definition (consecutive pose differences): 0.05–0.1% of the full-scale range, compared with the 0.3% we used. Both differences may have reduced DSD’s performance in our evaluation; their effect on our tasks was not measured.