To enhance accuracy of robot state estimation, perception-aware (or active sensing) methods seek trajectories that minimize uncertainty. To this aim, one possibility is to seek trajectories that minimize the final covariance of an extended Kalman filter (EKF), w.r.t. its control inputs over a given horizon. However, this can be computationally demanding. In this article, we derive novel backpropagation analytical formulas for the derivatives of the final covariance of an EKF w.r.t. its inputs. We then leverage the obtained gradients as an enabling technology to derive perception-aware optimal motion plans. Simulations validate the approach, showcasing improvements in both estimation accuracy and execution time. Experimental results on a real large ground vehicle also support the method.

翻译：为提升机器人状态估计精度，感知感知（或称主动感知）方法通过寻找使不确定性最小化的轨迹来实现这一目标。为此，一种可行方案是寻求在给定时间范围内，使扩展卡尔曼滤波（EKF）最终协方差相对于其控制输入最小化的轨迹。然而，这可能会带来巨大的计算负担。本文推导了EKF最终协方差相对于其输入的全新反向传播解析导数公式，进而利用所获梯度作为关键技术，生成感知感知的最优运动规划。仿真结果验证了该方法在提升估计精度与执行时间方面的有效性，真实大型地面车辆的实验结果也进一步支持了该方法的可行性。