Created by Emanuel Gollob, disarming is a series of performative explorations of the relation between a detached robotic arm, its artificial environment, and its human observers. Each iteration challenges the industrial robotic arm to adapt to its surroundings, continuously learning and unlearning, using the same Reinforcement Learning (RL) algorithm.
Locomotion can be seen as a primal (post-birth) instinct and ultimate act of independence. A robotic limb, somehow detached from a human-constructed technological body, tries to find concepts for advancing movements even though it initially wasn’t made for locomotion – vulnerable yet determined. Parallel to a familiar dystopian plot of technological autonomy and the feelings going with it, witnessing these first clumsy tries may awaken compassion or even a certain emotional bond.
Emanuel Gollob
In disarming II, the detached industrial robotic arm continuously tries to learn how to move forward from the status quo in the human-robot-environment relationship based on stereophonic listening and torque feedback. Dasha Bogdan, a dancer experienced in Butoh & contact dance, is asked to put herself in the situation of the robot and bodily imagine how she would re-learn body and movements new, based only on acoustic & haptic attentiveness. Mar Sala Romagosa, a sound artist & flute player, acoustically contributes her experience in attuning with instruments, improvisation based on listening, and musical references on bodies navigating in space and vice versa.
The third iteration is a 25-minute work of improvisational performative moments involving Wibke Storkan, the detached robotic arm ReBel and the audience of a medical robotics conference, among others.
In the latest iteration of disarming, a freely placed robotic arm durationally learns locomotion on a gym mat in relation to its embedded virtual concept of its body and environment. Exhibited as video installation, here physical acting is fused with digital observation and vice versa. In this instance, the same Reinforcement Learning (RL) algorithm now observes how likely recent social media postings mention “robot” and “arm” in the same post. In reference to this observation, the RL algorithm is set to continuously learn and unlearn how to edit the video scenes to decrease this human tendency.
The work is a response to the industry narrative that robots are becoming more and more autonomous with the help of machine learning. Emanuel reflects on the (human) ability to move independently as a core criterion of autonomy, or as a defining characteristic of being alive. Today, these definitions of autonomy and aliveness have evolved, but our inclination towards anthropomorphizing remains. In disarming, the learning is stretched and inefficient, creating a space to observe one’s projections and the multiplicity of relationships emerging and transforming between involved entities.
The project uses KUKA industrial robot arm, Reinforcement Learning, vvvv gamma, bunraku.xyz and KUKA’s Robot Sensor Interface.
Advice and Support Credits: Markus Krampl (video documentation), Amir Bastan (real-time robot control), Magdalena May (scenography), Creative Robotics (robotic hardware support), WRO ART Center team (curatorial and organizational support)
