Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
The Power of Predictions in Online Control
Authors: Chenkai Yu, Guanya Shi, Soon-Jo Chung, Yisong Yue, Adam Wierman
NeurIPS 2020 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To illustrate our theoretical results, we test MPC with different numbers of predictions in a Linear Quadratic (LQ) tracking problem, where the desired trajectory is given by: dt = 8 sin(t/3) cos(t/3) 8 sin(t/3) We consider following double integrator dynamics: pt+1 = pt + vt + ht, vt+1 = vt + ut + ηt, where pt R2 is the position, vt is the velocity, ut is the control, and ht, ηt U[ 1, 1]2 are i.i.d. noises. The objective is to minimize t=0 pt dt 2 + ut 2, where we let T = 200. This problem can be converted to the standard LQR with disturbance wt by letting xt = [ pt vt ] and wt = ht ηt and then using the reduction in the LQ tracking example in Section 2. Note that after the reduction, the disturbances are the combination of a deterministic trajectory and i.i.d. noises, which corresponds to the case discussed in Section 4. Figure 1 shows the tracking results with MPC using different numbers of predictions. We see that the regret exponentially decreases as the number of predictions increases, which is consistent with our theoretical results. |
| Researcher Affiliation | Academia | Chenkai Yu IIIS, Tsinghua University EMAIL Guanya Shi CMS, Caltech EMAIL Soon-Jo Chung CMS, GALCIT, JPL, Caltech EMAIL Yisong Yue CMS, Caltech EMAIL Adam Wierman CMS, Caltech EMAIL |
| Pseudocode | Yes | Algorithm 1: Model predictive control with k predictions Parameter: {A, B, Q, R} and Qf Rn n Input: x0, w0, . . . , wk 1 1 for t = 0 to T 1 do Input: xt, wt+k 1 // The controller now knows x0, . . . , xt, w0, . . . , wt+k 1 2 (ut, . . . , ut+k 1) = arg minu Pt+k 1 i=t (x i Qxi + u i Rui) + x t+k Qf xt+k subject to xi+1 = Axi + Bui + wi for i = t, . . . , t + k 1 Output: ut |
| Open Source Code | No | The paper does not include any statements about releasing source code for the described methodology or links to a code repository. |
| Open Datasets | No | The numerical experiments in Section 6 describe a synthetic simulation setup for an LQ tracking problem with specified dynamics and disturbances. It does not use a publicly available or open dataset with access information (link, DOI, formal citation). |
| Dataset Splits | No | The paper describes a numerical simulation setup in Section 6, not a traditional machine learning experiment with training, validation, and test dataset splits. |
| Hardware Specification | No | The paper does not specify any hardware details (e.g., CPU, GPU models, or cloud resources) used for running the numerical experiments. |
| Software Dependencies | No | The paper does not list any specific software dependencies with version numbers for its implementation or experiments. |
| Experiment Setup | Yes | To illustrate our theoretical results, we test MPC with different numbers of predictions in a Linear Quadratic (LQ) tracking problem, where the desired trajectory is given by: dt = 8 sin(t/3) cos(t/3) 8 sin(t/3) We consider following double integrator dynamics: pt+1 = pt + vt + ht, vt+1 = vt + ut + ηt, where pt R2 is the position, vt is the velocity, ut is the control, and ht, ηt U[ 1, 1]2 are i.i.d. noises. The objective is to minimize t=0 pt dt 2 + ut 2, where we let T = 200. |