Driving · 2026-10-01
SplineWAM: Adaptive Action Horizons for World Action Models via B-Spline Representations
Jun Guo, Xiaoshen Han, Qiwei Li, Nan Sun, Peiyan Li, Heyun Wang, Hang Lai, Weinan Zhang, Xinghang Li, Huaping Liu
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
SplineWAM replaces fixed-length action chunking in world action models with a fixed-size window of adaptive cubic B-spline parameters. The predicted knot times determine the decoded action duration and the interval until the next policy call. For asynchronous deployment, JP-RTC enforces continuity on decoded actions via a decoder Jacobian pullback. The authors report higher success with fewer policy calls on LIBERO-Plus and RoboCasa, and real-robot gains/match on three bimanual tasks under asynchronous execution.
Why it matters
The authors target an inefficiency of WAM action chunking: fixed computational/horizon allocation in time, which wastes calls over free-space and replans too slowly for contact-rich manipulation. By encoding an execution schedule directly in an action representation (knot times), and by enabling asynchronous consistency on executed actions (JP-RTC), SplineWAM aims to improve throughput (fewer policy calls) without reducing success.
Method
- Predict a fixed-size window of cubic B-spline parameters in the action branch of a WAM; knot times are fitted adaptively (error-driven) so decoding yields variable temporal resolution and duration.
- Align video supervision to fitted knot times (knot-grid alignment) instead of sampling on a uniform temporal grid.
- Deploy asynchronously with JP-RTC: impose continuity on decoded raw actions (not spline parameters) and pull corrections back through the decoder’s local Jacobian to match already-committed action prefixes.
Limitation
The authors state their spline actions are not generally as latency-tolerant as per-timestep chunks; JP-RTC recovers deficits opened by naive spline-space chunking but there remains a gap to synchronous execution (larger on LIBERO-Plus than on RoboCasa). They also note one fitting tolerance per dataset (chosen per benchmark and held fixed).
Abstract (from arXiv)
World action models (WAMs) are large embodied policies that jointly predict future video and the actions to execute, emitting a fixed-length action chunk per inference call. Such a policy allocates its computational budget uniformly in time, unable to execute for longer over free-space motion or to spend more inference on contact-rich manipulation, which limits the throughput a WAM can reach when served in the cloud. We present SplineWAM, which adaptively compresses the action trajectory into a fixed-size window of cubic B-spline parameters, fitting the knot times to the characteristics of the motion. One parameter budget then decodes into chunks of varying temporal resolution and duration, and both the executed span and the interval until the next policy call follow from the prediction itself. Aligning the video supervision to the fitted knot times of the demonstration rather than to a uniform grid concentrates the supervised frames where the action trajectory is complex. For asynchronous deployment we introduce Jacobian-Pullback Real-Time Chunking (JP-RTC), which imposes chunk continuity on the decoded raw actions the robot executes rather than on the spline parameters, and corrects the parameters through the decoder so that the executed prefix agrees with the actions already committed. On LIBERO-Plus and RoboCasa, SplineWAM improves success rate over an action chunking WAM by $8.2$ and $4.4$ points while cutting policy calls per episode by 22% and 26%. On three bimanual real-robot tasks under asynchronous execution, it leads or matches the baseline while decoding 1.2 to 1.6 times as much executed motion per call.
Related papers
- Staircase Policy: Streaming Inference for World-Action Models with Large Action Chunks
- {\Delta}WAM: Distilling Action Tangent Fields into World Action Models
- StreamingWAM: Streaming World Action Modeling via Chunk-Wise Staircase Flow Matching
- Long-WAM: Scaling the Context of World-Action Models
- Box-WAM: Box Trajectories as Compact State Representations for World Action Models
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-09-30 · Indexed by Pub-AI: 2026-10-01 UTC
SplineWAM: Adaptive Action Horizons for World Action Models via B-Spline Representations
Preprint posted: 2026-09-30 · Original record ↗
- arxiv
- 2609.39873
Explicit source versions (1)
- v1 · 2026-09-30
- Preprint available
Source date: 2026-09-30 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
World action models (WAMs) are large embodied policies that jointly predict future video and the actions to execute, emitting a fixed-length action chunk per inference call. Such a policy allocates its computational budget uniformly in time, unable to execute for longer over free-space motion or to spend more inference on contact-rich manipulation, which limits the throughput a WAM can reach when served in the cloud. We present SplineWAM, which adaptively compresses the action trajectory into a fixed-size window of cubic B-spline parameters, fitting the knot times to the characteristics of the motion. One parameter budget then decodes into chunks of varying temporal resolution and duration, and both the executed span and the interval until the next policy call follow from the prediction itself. Aligning the video supervision to the fitted knot times of the demonstration rather than to a uniform grid concentrates the supervised frames where the action trajectory is complex. For asynchronous deployment we introduce Jacobian-Pullback Real-Time Chunking (JP-RTC), which imposes chunk continuity on the decoded raw actions the robot executes rather than on the spline parameters, and corrects the parameters through the decoder so that the executed prefix agrees with the actions already committed. On LIBERO-Plus and RoboCasa, SplineWAM improves success rate over an action chunking WAM by $8.2$ and $4.4$ points while cutting policy calls per episode by 22% and 26%. On three bimanual real-robot tasks under asynchronous execution, it leads or matches the baseline while decoding 1.2 to 1.6 times as much executed motion per call.