Robot Trajectron: Trajectory Prediction-based Shared Control for Robot Manipulation

TL;DR

Robot Trajectron (RT) addresses trajectory-based intent estimation for shared-control manipulation by predicting end-effector motion from short history of position, velocity, and acceleration using a CVAE-style probabilistic model with latent dynamics. It couples RT with a novel shared-control scheme based on Artificial Potential Fields that integrates RT's goal distribution and predicted trajectory, moderated by an agreement mechanism to handle user intent changes. The approach achieves accurate short-horizon trajectory prediction, enables robust intent estimation in 3D tabletop tasks, and demonstrates improved efficiency and smoother motion in both simulation and real-robot teleoperation, with performance competitive to state-of-the-art IOC baselines. The work provides data and code for public use, highlighting practical impact for reducing operator cognitive load in assisted manipulation tasks and enabling responsive adaptation to changing user intent.

Abstract

We address the problem of (a) predicting the trajectory of an arm reaching motion, based on a few seconds of the motion's onset, and (b) leveraging this predictor to facilitate shared-control manipulation tasks, easing the cognitive load of the operator by assisting them in their anticipated direction of motion. Our novel intent estimator, dubbed the \emph{Robot Trajectron} (RT), produces a probabilistic representation of the robot's anticipated trajectory based on its recent position, velocity and acceleration history. Taking arm dynamics into account allows RT to capture the operator's intent better than other SOTA models that only use the arm's position, making it particularly well-suited to assist in tasks where the operator's intent is susceptible to change. We derive a novel shared-control solution that combines RT's predictive capacity to a representation of the locations of potential reaching targets. Our experiments demonstrate RT's effectiveness in both intent estimation and shared-control tasks. We will make the code and data supporting our experiments publicly available at https://github.com/mousecpn/Robot-Trajectron.git.

Figures (6)

• Figure 1: The architecture of the Robot Trajectron. The red lines denote the train-only operations, while the green lines denote the predict-only operations. See text for details.
• Figure 2: Overview of the proposed shared control. $\bm{v}_g$ denotes the velocity towards the most likely object, given the predicted trajectory. $\bm{v}_{\textnormal{tr}}$ denotes the velocity along the trajectory predicted by RT. $\bm{v}_u$ denotes noisy user command. See text for details.
• Figure 3: Visualization of RT. The most likely trajectory and the 2D table GMMs are shown. The green line denotes the past trajectory, while the red line denotes the predicted trajectory.
• Figure 4: (a) The set up of the shared autonomy experiment. (b) A demonstration assisted by the proposed method. (c) The weight change in Fig. b. (d) The distribution change of each goal in Fig. b.
• Figure 5: User demonstrations. The blue lines denote the trajectories fully controlled by the user. The red and green lines denote the trajectories assisted by MaxEnt IOC and Robot Trajectron, respectively. As discussed in the text, the trajectories guided by RT to reach the object are straighter and smoother.