Conditional Neural Movement Primitives

Abstract

Conditional Neural Movement Primitives (CNMPs) is a learning from demonstration framework that is designed as a robotic movement learning and generation system built on top of a recent deep neural architecture, namely Conditional Neural Processes (CNPs). Based on CNPs, CNMPs extract the prior knowledge directly from the training data by sampling observations from it, and uses it to predict a conditional distribution over any other target points. CNMPs specifically learns complex temporal multi-modal sensorimotor relations in connection with external parameters and goals; produces movement trajectories in joint or task space; and executes these trajectories through a high-level feedback control loop. Conditioned with an external goal that is encoded in the sensorimotor space of the robot, predicted sensorimotor trajectory that is expected to be observed during the successful execution of the task is generated by the CNMP, and the corresponding motor commands are executed. In order to detect and react to unexpected events during action execution, CNMP is further conditioned with the actual sensor readings in each time-step. Through simulations and real robot experiments, we showed that CNMPs can learn the nonlinear relations between low-dimensional parameter spaces and complex movement trajectories from few demonstrations; and they can also model the associations between high-dimensional sensorimotor spaces and complex motions using large number of demonstrations. The experiments further showed that even the task parameters were not explicitly provided to the system, the robot could learn their influence by associating the learned sensorimotor representations with the movement trajectories. The robot, for example, learned the influence of object weights and shapes through exploiting its sensorimotor space that includes proprioception and force measurements; and be able to change the movement trajectory on the fly when one of these factors were changed through external intervention.

Motion Adaptation Based on Learning the Manifold of Task and Dynamic Movement Primitive Parameters

“…49,50 In such cases, the number of task parameters is large; therefore, kernel estimation may not be suitable and using a DNN may be required. Seker et al 3 suggest an alternative primitive representation based on the conditional neural processes DNN architecture. Augmenting such a representation with a learning and generation system facilitates encoding complex temporal multi-modal sensorimotor relations in connection with complex task constraints.…”

“…1 In light of this, several forms of motor primitives have been suggested for robot motion planning and control. 2,3 When motion is based on motor primitives, learning to perform a task is related to learning the suitable motion primitive parameters. Similarly, performance adaptation and improvement during runtime are related to adaptation of the motion primitive parameter values.…”

Analyzing Neural Jacobian Methods in Applications of Visual Servoing and Kinematic Control

“…We include ProMPs because it is the foundational framework that motivates our work. CNMP can be viewed as a version of our model with fixed variance in the latent distribution, but that predicts output variance σ y (z, x) as done in previous work [21]. VAE-CNMP is a variation of our model without Bayesian aggregation, the via-point, and context variable independence assumption, but still uses the Isotropic Gaussian as the prior in the KL divergence regularization.…”

“…The pivotal work of Seker et al [21] overcomes the limitations of previous work by proposing a continuoustime, non-linear representation of motor skills. These motor primitives utilizes Conditional Neural Processes (CNPs) [7], [8], an encoder-decoder model that allows aggregating multiple input-output pairs to form a latent representation of a function.…”

One-Shot Multi-Path Planning Using Fully Convolutional Networks in a Comparison to Other Algorithms

“…Recently, several path planning methods have been proposed using shallow (De Momi et al, 2016) or deep network approaches such as deep multi-layer perceptrons (DMLP; Qureshi et al, 2019), long short-term memory (LSTM) networks (Bency et al, 2019), and deep reinforcement learning (deep-RL; Tai et al, 2017;Panov et al, 2018), or mixed approaches that can be used for robot movement generation (Seker et al, 2019). All these methods plan paths iteratively by predicting the next state or next action (in case of RL) based on the current state, environment configuration, and the target position until the goal is reached.…”