Chronos: A Physics-Informed Full-History Framework for Non-Markovian Long-Horizon Manipulation

2026-06-29

AI Summary

This work addresses the failure of Markovian policies in long-horizon, memory-dependent robotic tasks by proposing a non-Markovian policy framework that explicitly models the full observation history as the policy’s latent state rather than auxiliary context. The approach encodes historical observations into a token sequence temporally aligned with physical time and leverages a selective state-space model, implicit maximum likelihood estimation, a second-order Schrödinger bridge, and a multimodal action prior to generate smooth, dynamics-aware actions. Evaluated on RMBench, the method achieves an average success rate of 73.6%, outperforming Markovian baselines by 62.4 percentage points. In real-world dual-arm experiments, it attains an overall success rate of 78% (72% on memory-intensive tasks) while reducing parameter count by 10–30×.

Abstract

General-purpose robot policies should be modeled as dynamical systems, yet many VLA and generative imitation policies still rely on present observations or short windows. This Markovian shortcut fails in memory-dependent manipulation: identical observations can demand different actions after different histories. We present Chronos, a physics-informed full-history framework for non-Markovian long-horizon manipulation. The key idea is to elevate observation history from auxiliary context to the latent state of the policy dynamics. At each physical control step, Chronos forms one state-representative token by fusing observation and proprioception, so the token sequence is aligned one-to-one with physical time. A selective state space model propagates this causal historical state, which conditions a multimodal coarse action prior through implicit maximum likelihood estimation (IMLE). This prior is then refined by a second-order Schrodinger-inspired bridge that predicts acceleration fields, yielding smoother and more physically grounded robot motion. Across 16 simulated tasks and 4 real-world experiments, Chronos is evaluated on precision insertion, general manipulation, and memory-dependent long-horizon control. On RMBench, where success requires remembering task phase, Chronos achieves 73.6% average success, outperforming Markovian VLA baseline pi0.5 by +62.4 percentage points, a 6.6x relative gain, while using 10x fewer parameters. It also surpasses the memory VLA Mem-0 by 22.8 points while using over 30x fewer parameters. In real-world dual-arm experiments using a single RGB camera, Chronos achieves 78% average success over four tasks, including 72% on the three memory-dependent tasks, whereas pi0.5 achieves 7% overall and 0% on the memory-dependent subset. These results suggest that history should not be treated as auxiliary context, but as the latent state of the manipulation policy.

Problem

• non-Markovian
• long-horizon manipulation
• memory-dependent control
• robot policy
• history dependence

Innovation

• non-Markovian
• full-history modeling
• physics-informed policy
• state space model
• implicit maximum likelihood estimation