The field of edge intelligence and autonomous systems is rapidly advancing, with a focus on optimizing hardware efficiency, developing scalable training methodologies, and improving performance in dynamic environments. Recent developments have seen the proposal of novel multiplier designs, reinforcement learning frameworks, and approximate processing architectures, all aimed at enhancing computational efficiency and robustness in resource-constrained environments. Notably, innovative approaches such as self-play mechanisms and curriculum-based iterative self-play have shown significant promise in developing robust and adaptive strategies for autonomous agents.

Some noteworthy papers in this regard include: SPIRAL, which introduces a self-play incremental racing algorithm for learning in multi-drone competitions, demonstrating significant advantages in developing sophisticated cooperative multi-drone racing strategies. CRUISE, which presents a curriculum-based iterative self-play framework for scalable multi-drone racing, achieving nearly double the mean racing speed of a state-of-the-art game-theoretic planner. RAMAN, which proposes a resource-efficient approximate posit-based Multiply-Accumulate architecture, achieving up to 46% in LUT savings and 35.66% area reduction over the baseline design. Accelerating Real-World Overtaking in F1TENTH Racing, which presents a novel racing and overtaking agent capable of learning to reliably navigate a track and overtake opponents in both simulation and reality, demonstrating an overtaking rate of 87%.