Aniket Bera @ - 3rd+ + Follow == 8 Associate Professor in Computer Science at Purdue Universit... with 4 others + Oct 24 - Edited True Embodied Al must run on the chip, not in the cloud. We developed every layer of this autonomous drone from scratch here in the IDEAS Lab at Purdue: from CAD design and 3D-printed carbon fiber airframes to the custom research pipelines and on-chip software stack to push high-speed, on-device intelligence into one of the most adversarial environments possible: an unmapped, GPS- denied forest. Tracking a running target under a dense canopy means the entire autonomy pipeline: visual-inertial odometry, local occupancy mapping, dynamic occlusion prediction, and perception-aware trajectory optimization- must execute within strict SWaP-C and thermal envelopes on embedded silicon. When banking hard to dodge a thin branch perturbs the sensor frustum, we cannot afford offboard latency; closing the loop from pixels to motor actuation in milliseconds right on the edge is what keeps the drone alive and the runner in frame. EDIT: This work involves many open-source software tools, but the key novel planning and perception algorithms have been developed in-house and are under review; a preprint will be available soon. #Purdue #EmbodiedAl #EdgeAl #0nChipAl #Robotics #MotionPlanning #UAV SSEe NER EE 00 Sel CC == & AREER 3S he ams Ev EER SN rh fhe i ioe ys IER 0 Sm CE ERE Te A he HETETER DW Spa LR N Sas; Ant 8 Ii iy ee Sia \ { Abed, le ir, | / d oo AN / | J p 3 d ld : - Gir) {- 1 i