Figure 4：Under nonlinear TPPP operation, autonomous driving task is performed. TA-RNN utilizes temporally learnt parameter, allowing for swifter control, significantly improves trajectory prediction of autonomous driving. (IMAGE)

Light Publishing Center, Changchun Institute of Optics, Fine Mechanics And Physics, CAS

Figure 4：Under nonlinear TPPP operation, autonomous driving task is performed. TA-RNN utilizes temporally learnt parameter, allowing for swifter control, significantly improves trajectory prediction of autonomous driving.

Caption

Figure 4：Under nonlinear TPPP operation, autonomous driving task is performed. TA-RNN utilizes temporally learnt parameter, allowing for swifter control, significantly improves trajectory prediction of autonomous driving. (a) Schematic overview of a time-adaptive RNN implemented on TPPP hardware for autonomous driving control. Sensor streams, including vehicle speed and front, left and right obstacle distances, are processed by a TA-RNN mapped onto TPPP hardware. The dynamic kernel updates the input-to-hidden and hidden-to-hidden weight matrices over time, enabling real-time prediction of throttle and steering commands. (b) Multi-layer TA-RNN with temporal cascading; at each time step, blocks denote combined weight matrices (Wih and Whh), with sensor inputs (v, dF, dL, dR) and control outputs (steering, throttle). (c) Mean encoding accuracy over time steps for Sih. (d) Mean encoding accuracy over time steps for Shh. (e) Steering command and (f) throttle command computed by the TPPP and the corresponding ground-truth values. (h) Driving trajectory comparison between ground truth and predictions from a conventional RNN and a TA-RNN

Credit

Qixiang Cheng et al.

Usage Restrictions

Credit must be given to the creator.

License

CC BY

Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.