Predictive Linear Online Tracking for Unknown Targets
Authors: Anastasios Tsiamis, Aren Karapetyan, Yueshan Li, Efe C. Balta, John Lygeros
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We implement PLOT on a real quadrotor and provide open-source software, thus, showcasing one of the first successful applications of online control methods on real hardware. and 5. Simulations and Experimental Validation |
| Researcher Affiliation | Collaboration | 1Automatic Control Laboratory, ETH Z urich, Z urich, Switzerland 2Institute for Dynamic Systems and Control, ETH Z urich, Z urich, Switzerland 3Control and Automation Group, inspire AG, Z urich, Switzerland. |
| Pseudocode | Yes | Algorithm 1 RLS for k-step-ahead prediction and Algorithm 2 PLOT: Predictive Linear Online Tracking |
| Open Source Code | Yes | The code for both the simulation and hardware experiments can be accessed from https://gitlab.nccr-automation.ch/akarapetyan/plot. |
| Open Datasets | No | The paper uses data from simulations of a linearized quadrotor model and experiments on a real Crazyflie quadrotor. It generates virtual reference trajectories online, but does not refer to a publicly available or open dataset with access information or a citation. |
| Dataset Splits | No | The paper focuses on online tracking and adaptation, where models learn sequentially from incoming data rather than being trained on pre-split datasets. Therefore, explicit training, validation, or test dataset splits are not applicable or mentioned. |
| Hardware Specification | Yes | We implement PLOT on a real quadrotor and provide open-source software, thus, showcasing one of the first successful applications of online control methods on real hardware. and We validate the proposed algorithm on the Crazyflie 2.1 quadrotor. and The Crazyflie quadrotors are equipped with a 3-axis accelerometer and gyroscope for onboard angular-rate control and 3 retro-reflective markers, as shown in Figure 20. These markers are detected by 5 VICON cameras mounted around the flying arena, shown in Figure 21, and the VICON system provides a spatial position estimate of the drone in its precalibrated reference frame of reference. and The control inputs are sent to the quadrotor through a radio communicator from a centralized computer that receives state measurements through a local area network connection to the motion capture system, and runs the online controller. |
| Software Dependencies | No | The paper mentions 'Robot Operating System (ROS)' but does not provide specific version numbers for ROS or any other software dependencies. The general statement about ROS is insufficient for reproducibility without version details. |
| Experiment Setup | Yes | For all examples, we take the sampling time to be Ts = 0.1 seconds and the LQR cost matrices fixed at Q := diag(80, 80, 80, 10, 10, 10, 0.01, 0.01, 0.1), and R := diag(0.7, 2.5, 2.5, 2.5). and For the PLOT policy, we take p = 1 for all experiments, and the projection step 8 in the RLS Algorithm 1 is executed with a quadratic program solver, solving the following problem and for all k = 1, . . . , W, the k learners are initialized as follows: ˆSj+k|j = I6, vj+k|j = 06 and Pj|j k = 10 4 I7, for all j = 1, . . . , k 2. and We implement PLOT for the linearized model of the drones derived in Appendix F with a fixed prediction horizon of W = 5 and a forgetting factor of γ = 0.8. |