Emergent Hand Morphology and Control from Optimizing Robust Grasps of Diverse Objects

Abstract

Evolution in nature illustrates that the creatures' biological structure and their sensorimotor skills adapt to the environmental changes for survival. Likewise, the ability to morph and acquire new skills can facilitate an embodied agent to solve tasks of varying complexities. In this work, we introduce a data-driven approach where effective hand designs naturally emerge for the purpose of grasping diverse objects. Jointly optimizing morphology and control imposes computational challenges since it requires constant evaluation of a black-box function that measures the performance of a combination of embodiment and behavior. We develop a novel Bayesian Optimization algorithm that efficiently co-designs the morphology and grasping skills jointly through learned latentspace representations. We design the grasping tasks based on a taxonomy of three human grasp types: power grasp, pinch grasp, and lateral grasp. Through experimentation and comparative study, we demonstrate the effectiveness of our approach in discovering robust and cost-efficient hand morphologies for grasping novel objects. https://xinleipan.github.io/ emergent_morphology/ 1 NVIDIA,

Parameter Identification and Motion Control for Articulated Rigid Body Robots Using Differentiable Position-based Dynamics

“…It is challenging to ensure the design parameters without incorporating motion control constraints (such as d.o.f, work-space, etc.). In literature, the parameterization methods based on primitive shapes have been proposed using gradient-based [27] and gradientfree [28] approaches for optimizations of design. However, these primitive shapes are limited by simple geometrical features.…”

“…τ net (28) The PD control terms τ pd includes damping term D and stiffness term K. Dynamics models could be used for calculating net torque τ net includes inertial, Coriolis, and gravitational elements, which would be predicted based on optimization in our differentiable simulation framework with a given initial state and a final state.…”

REvolveR: Continuous Evolutionary Models for Robot-to-robot Policy Transfer

“…Vanilla GNNs (Scarselli et al, 2009) have been used to control diverse robot morphologies in NerveNet (Wang et al, 2018) to control different robots obtained by growing the limbs within topology (Wang et al, 2019b;Hejna III et al, 2021) or across topology (Gupta et al, 2021). Learning-driven evolution could be used to improve the design as well of the agent (Cheney et al, 2014;Ha et al, 2017;Ha, 2018;Schaff et al, 2018;Pan et al, 2021). Similarly, one could also evolve the environment itself too (Wang et al, 2019a).…”

N-LIMB: Neural Limb Optimization for Efficient Morphological Design

“…A common approach to optimizing over discrete designs is to perform evolutionary search (ES) over a population of candidate designs [5,6,19,20,21,7,22,23,24,25,8,26,9,27]. However, ES is prone to local minima and is sample-inefficient, in part due to the need to maintain separate control policies for each candidate design in the population.…”