In context
In early 2025, autonomous robots still struggled with learning in unknown environments, often limited by pre-collected training data and the disconnect between perception and action. This paper, published in the Chinese journal Robot, addressed that gap by proposing a joint learning framework that unifies sensorimotor learning, a key challenge for industrial robots operating in dynamic settings.
What was reported
Researchers from Chongqing University introduced a sensorimotor joint learning approach combining human simulated intelligent schema theory with active inference. The system is structured into three schemas: perceptual, associative, and motor. The perceptual schema processes raw sensor data (e.g., images) using a PP-YOLO-based detection model, while the associative schema links perception to action via a generative model and an inference model. A cognitive module, inspired by Piaget's assimilation and accommodation, enables the robot to learn from unfamiliar inputs and generate new training data autonomously.
The joint learning algorithm coordinates these modules, allowing the robot to actively explore and refine both its perception and motion. In experiments, the method improved perception accuracy by 22.32% in simulation and 12.00% in physical tests, while significantly reducing time to reach target positions and producing smoother trajectories.
The approach addresses the problem of partial observability by using variational Bayesian inference to approximate posterior probabilities, enabling action inference even when sensory data is incomplete.
Why it mattered
This work demonstrated a systematic way to couple perception and action in robot learning, moving beyond separate training of perception and control. For industrial automation, such joint learning could enable robots to adapt to new environments with less human intervention, improving autonomy and efficiency in tasks like inspection or manipulation.
"The joint learning method can enhance the robot's cognition and self-learning abilities, as well as can autonomously optimize recognition accuracy and movement trajectory."
Source: 《机器人》期刊 (robot.sia.cn) · Published 2025-03-15 · “基于主动推理与仿人智能图式的感知运动联合学习方法”