Tamim Asfour, a humanoid robotics researcher at KIT in Karlsruhe, discusses motion intelligence and versatile robots for household, industrial, and assistive tasks at the Latin American Summer School on Cognitive Robotics.
Asfour describes more than 20 years of work on robots for daily activities, including the ARMAR series and systems for industrial collaboration. His group combines hardware engineering with learning from human demonstrations so robots can handle varied tasks rather than a single specialized job.
His team connects planning and control through memory, including episodic records of images, robot poses, object locations, actions, and timing. These records can support time-aware plans and answers to task-related questions, such as where the robot last saw a red cup. For verbal instructions, learned 3D spatial relations help translate phrases like placing mustard on a rusk into candidate positions and robot actions.
Asfour explains how robots learn actions such as wiping by capturing both movement and constraints like surface contact. Via-point movement primitives let a robot adapt a learned reach around an obstacle without relearning the entire action. His group also collected 42 hours of labeled whole-body human motion data and developed pose and bimanual-action taxonomies to organize demonstrations for robot use.
In warehouse maintenance, ARMAR-6 identifies when a technician may need help using visual, force, or speech cues, and the team tested the robot with technicians for six weeks. Other applications include grasping unfamiliar objects for nuclear decontamination, prosthetic hands that select among 13 grasp categories, and personalized exoskeletons. Asfour argues that useful robotics must connect perception to action, adapt through real-world interaction, and account for privacy, safety, trust, and social acceptance.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.
Tamim Asfour, professor at Karlsruhe Institute of Technology, presents research on robotic manipulation at ICRA 2025, following an introduction by Kevin Lynch of Northwestern University.
