Pay Attention! - Robustifying a Deep Visuomotor Policy Through Task-Focused Visual Attention

Abstract

Several recent studies have demonstrated the promise of deep visuomotor policies for robot manipulator control. Despite impressive progress, these systems are known to be vulnerable to physical disturbances, such as accidental or adversarial bumps that make them drop the manipulated object. They also tend to be distracted by visual disturbances such as objects moving in the robot's field of view, even if the disturbance does not physically prevent the execution of the task. In this paper, we propose an approach for augmenting a deep visuomotor policy trained through demonstrations with Task Focused visual Attention (TFA). The manipulation task is specified with a natural language text such as "move the red bowl to the left". This allows the visual attention component to concentrate on the current object that the robot needs to manipulate. We show that even in benign environments, the TFA allows the policy to consistently outperform a variant with no attention mechanism. More importantly, the new policy is significantly more robust: it regularly recovers from severe physical disturbances (such as bumps causing it to drop the object) from which the baseline policy, i.e. with no visual attention, almost never recovers. In addition, we show that the proposed policy performs correctly in the presence of a wide class of visual disturbances, exhibiting a behavior reminiscent of human selective visual attention experiments. Our proposed approach consists of a VAE-GAN network which encodes the visual input and feeds it to a Motor network that moves the robot joints. Also, our approach benefits from a teacher network for the TFA that leverages textual input command to robustify the visual encoder against various types of disturbances.

Object Detection-Based One-Shot Imitation Learning with an RGB-D Camera

“…As mentioned above, the robustness and generalization of these end-to-end learning-based methods are limited by the range of experience in the training phase. Abolghasemi et al applied a task-focused visual attention mechanism in the end-to-end IL framework to enhance the robustness of visuomotor manipulation skills [22]. James et al trained a visuomotor skill for pick-and-place task in simulation with domain randomization and transferred the visuomotor skill to the real world without any fine-tuning [3].…”

Composable Instructions and Prospection Guided Visuomotor Control for Robotic Manipulation

“…It was also used for motion switching in multi-step task such as cloth-folding [45]. Abolghasemi et al improved the robustness of this multi-task approach via task-focused visual attention [15]. Hausman et al also combined an embedding task space with RL to learn transferable robot skills [46].…”

“… res is the deviation of the observation o and the restored o , , which is expressed with the square of Euclidean Norm and  VAE is the total loss of VAE. Inspired by the work of Abolghasemi [15], another loss term that is the deviation of the reconstructed images of encoded vector is added to enhance coding performance.…”

“…where a k is optimal action of the observation o based on the specified instructions I in demonstrations . However, this sequential decision-making problem of robotic manipulation is not ideally Markov decision process (MDP), and the action depends not only on the current state or observation but also on the historical trajectory of state or observation, especially in complex situations [14,15].…”

“…In the past few years, robotics researchers and artificial intelligence (AI) scientists are actively exploring the visuomotor control approach for robotic manipulation to expand its application field, enhance its robustness, accelerate its learning speed and so on [12,13]. Some studies also have been done through this learning-based end-to-end visuomotor control for multi-task or multi-step situations [14,15]. Previous research has validated that end-to-end visuomotor control for multi-task manipulation is more sample-efficient and robust compared to methods applied to single tasks [14].…”