Robots tap human instinct to prevent costly mistakes in real time

Researchers at Oklahoma State University are developing a neuroadaptive control system that enables robots to detect and respond to human instinctive error recognition in real time. By using brain-computer interfaces with wearable EEG caps, the system captures error-related potentials (ErrPs)—electrical signals generated in the anterior cingulate cortex when a person perceives a mistake. These signals occur faster than physical reactions, allowing robots to slow down, stop, or return control to the human operator within milliseconds, thereby preventing costly or dangerous errors before they escalate. The system addresses limitations in current teleoperation, where humans remotely guide robots but often react too slowly to sudden failures. An adaptive decoding model personalizes the detection of brain signals to individual users, reducing setup time and increasing practicality. Safety is ensured through Signal Temporal Logic, which enforces strict behavioral constraints on the robot’s actions even when responding to brain signals. Tested with NVIDIA Isaac Lab and RTX PRO 6000 GPUs, this technology has potential applications beyond industrial settings, including healthcare