Online Intention Prediction via Control-Informed Learning

Tianyu Zhou; Zihao Liang; Zehui Lu; Shaoshuai Mou

cs.ROApr 10, 2026

Online Intention Prediction via Control-Informed Learning

Authors: Tianyu Zhou, Zihao Liang, Zehui Lu, Shaoshuai Mou

Organizations: School of Aeronautics and Astronautics, Purdue University, IN 47907, USA · independent researcher

Abstract

This paper presents an online intention prediction framework for estimating the goal state of autonomous systems in real time, even when intention is time-varying, and system dynamics or objectives include unknown parameters. The problem is formulated as an inverse optimal control / inverse reinforcement learning task, with the intention treated as a parameter in the objective. A shifting horizon strategy discounts outdated information, while online control-informed learning enables efficient gradient computation and online parameter updates. Simulations under varying noise levels and hardware experiments on a quadrotor drone demonstrate that the proposed approach achieves accurate, adaptive intention prediction in complex environments.

Figures & tables

Fig. 1 : A robot with an unknown intention was observed while executing its trajectory autonomously, during which the proposed algorithm predicted its intention in real time.

Fig. 2 : Strategy of shifting horizon in prediction.

Panel of Figure 3 — Fig. 3 : Intention prediction for quadrotor. (No goal change)

Panel of Figure 4 — Fig. 4 : Online intention prediction for quadrotor. (Two goal switches. Vertical dashed lines in (a) indicate goal changes.)

Fig. 5 : Online intention prediction with three goal switches, i.e. at $t=50$ , the system is at $\bar{\bm{x}}_{50}^{*}$ and switches its goal from $\bm{x}_{\mathrm{g},0}^{*}$ to $\bm{x}_{\mathrm{g},50}^{*}$ (goal $\bm{x}_{\mathrm{g},0}^{*}$ , $\cdots$ , $\bm{x}_{\mathrm{g},49}^{*}$ maintains the same).

Panel of Figure 6 — Fig. 6 : (a) Comparison of prediction loss over time between the proposed method and OCIL. $\Delta$ denotes the time interval between goal switches. (b) Comparison of average prediction loss between the proposed method and OCIL.

Panel of Figure 7 — Fig. 7 : Snapshots of the experiment for a quadrotor drone. Fig. 7(a) shows that at $t=35$ , the prediction indicates the quadrotor intends to fly in the positive $x$ -direction but does not yet reflect an intention to avoid the no-fly zone or accurately predict its true goal. By $t=60$ , as shown in Fig. 7(b) , the predicted goal shifts into the correct region as the quadrotor begins its second turn. As more measurements are collected, the prediction continues to improve and eventually converges to the ground truth, as illustrated in Fig. 7(c) . The prediction loss is included in Fig. 7(c) .

Online Learning

Online learning studies algorithms that update predictions or decisions from sequential feedback while controlling cumulative error or regret.

See all 132 papers

Neuro-Inspired Inverse Learning for Planning and Control
May 22, 2026Maryna Kapitonova, Tonio BallRobot Policy Learning Model-Based Planning
NeuroMentum AI · IMBIT, University of Freiburg, Germany
Self-Adaptive Learning and Model Predictive Control for Tracking Unknown Dynamics with No Regret
Jul 29, 2026Atharva Navsalkar, Hongyu Zhou, Vasileios TzoumasModel Predictive Control Dynamical Systems
Department of Aerospace Engineering, University of Michigan, Ann Arbor, MI 48109 USA
Adapting Neural Robot Dynamics on the Fly for Predictive Control
Apr 5, 2026Abdullah Altawaitan, Nikolay AtanasovRobot Dynamics Learning Robot Dynamics Identification
Department of Electrical and Computer Engineering, University of California San Diego, La Jolla, CA 92093, USA · Kuwait University

May 22, 2026J/K move · Enter open · S save

Neuro-Inspired Inverse Learning for Planning and Control

Maryna Kapitonova, Tonio Ball

NeuroMentum AI · IMBIT, University of Freiburg, Germany

Robot Policy Learning Model-Based Planning Model-Based RL

Open paper AskPDF arXiv

We present a neuro-inspired framework for embodied planning and control. Building on three principles that enable fast and highly effective goal-directed behavior in the mammalian brain - paired forward/inverse internal models, open-loop multi-step motor commands, and sequential, hierarchical organization of action - our Inverter framework uses learned components, trained end-to-end through Inverse Learning (IL) and supplemented where natural by analytic or algorithmic modules; we formalize IL and delineate it from supervised, reinforcement, and imitation learning. IL bridges Reinforcement Learning (RL)-style amortization, which runs in a single forward pass but emits only one action at a time, and Optimal Control (OC)-style sequence planning over whole trajectories, but with iterative test-time computation. Single Inverters or hierarchical n=2 Inverter stacks match or improve on offline-RL and diffusion-planner baselines on all 3 maze2d and 6 antmaze D4RL variants by an average of +24.2% (range -1.9% to +78.2%), at one-to-two orders of magnitude less inference compute time. Distinctively, optimizing through the Forward Model (FoM) over the entire T-step action sequence - rather than per step - lets Inverters produce smooth, goal-coherent, trajectory-wide structure and reach control policies closer to the analytic optimum than the policy underlying the training data itself. We also identify a failure mode of IL: FoM hacking under narrow training-data coverage, which we mitigate by using random training data with broader coverage. As an application example, a Pulse Inverter synthesizes arbitrary single-qubit quantum gates with fidelity matching the standard iterative numerical baseline (GRAPE), at more than 1000x lower per-gate compute time. In summary, we conclude that IL enables a versatile class of world-interfaces, especially for latency- and resource-critical embodied AI.

Abstract

Online Learning

Explore similar work

Neuro-Inspired Inverse Learning for Planning and Control

Self-Adaptive Learning and Model Predictive Control for Tracking Unknown Dynamics with No Regret

Adapting Neural Robot Dynamics on the Fly for Predictive Control