ETA-IK: Execution-Time-Aware Inverse Kinematics for Dual-Arm Systems

TL;DR

ETA-IK addresses dual-arm inverse kinematics under relative-pose constraints by directly minimizing execution time rather than relying on surrogate joint-distance metrics. It combines a relative TCP pose formulation with a neural, differentiable execution-time estimator trained on offline TOPPRA and TrajOpt data, enabling a differentiable, time-aware multi-objective IK solved in parallel with many seeds. The approach demonstrates significant time reductions on a UR5–KUKA IIWA system while preserving positioning accuracy, and includes collision-awareness through training data and objective terms. While effective, the method incurs additional neural-inference overhead requiring a capable GPU, with suggested future work to reduce latency and refine initial pose distributions.

Abstract

This paper presents ETA-IK, a novel Execution-Time-Aware Inverse Kinematics method tailored for dual-arm robotic systems. The primary goal is to optimize motion execution time by leveraging the redundancy of both arms, specifically in tasks where only the relative pose of the robots is constrained, such as dual-arm scanning of unknown objects. Unlike traditional inverse kinematics methods that use surrogate metrics such as joint configuration distance, our method incorporates direct motion execution time and implicit collisions into the optimization process, thereby finding target joints that allow subsequent trajectory generation to get more efficient and collision-free motion. A neural network based execution time approximator is employed to predict time-efficient joint configurations while accounting for potential collisions. Through experimental evaluation on a system composed of a UR5 and a KUKA iiwa robot, we demonstrate significant reductions in execution time. The proposed method outperforms conventional approaches, showing improved motion efficiency without sacrificing positioning accuracy. These results highlight the potential of ETA-IK to improve the performance of dual-arm systems in applications, where efficiency and safety are paramount.

Figures (5)

• Figure 1: Application and Motivation: Improving execution time for modeling radioactive unknown objects with two robot arms (top left). Given a set of relative poses representing the perception poses (bottom left), ETA-IK aims to use the redundant DoF to solve each single IK problem and find the optimal and collision-free joint configurations (top right), to accelerate the entire scanning process (bottom right).
• Figure 2: Scanning and modeling with two different manipulators: The KUKA iiwa robot picks up the target object (stone) and the UR5 robot carries a 3D lidar scanner. To get a complete and accurate model, both robots have to change the scanning pose through many relative poses that are projected as the best scanning perspective. This process needs to be accelerated by using our proposed approach.
• Figure 3: Pipeline of ETA-IK: given a desired perception pose, ETA-IK first generate a batch of initial configurations (yellow). A multilayer-perceptron time apporiximator are then integrated in a parallel optimization framework (blue). After $N$ iterations, the results is a batch of joint configurations. The best solution is selected according to different criteria (green).
• Figure 4: Execution time approximator is able to implicitly encode the self-collision or static environment information when the dataset considered collision avoidance. Considering the collision, the green trajectory is much longer than simply connecting the start and end configurations, shown in cyan.
• Figure 5: Execution time comparison between reference, method (C) and (G) for collision-free trajectory generation: The starting pose and the reference target shown are randomly generated. The starting joint configuration, reference target configuration, and the IK solution generated by method (C) and (G) in Tab. \ref{['tab:exp_2']} are illustrated with the corresponding position and velocity profile over time generated by TrajOpt.