Learning-based driving planners are usually trained and evaluated in open loop against logged trajectories. In closed loop, a trajectory with small displacement error can still stall the vehicle, steer it into a conflict with surrounding agents, or be executed with abrupt braking. We introduce Closed-Loop Refinement and Execution (CLRE), a hierarchical receding-horizon control framework designed to mitigate these failure modes while leaving the upstream planner frozen and adding no new learned model. The upper layer treats the nominal trajectory as a reference and solves a finite-horizon optimal control problem that trades route progress against interaction with predicted agents. Solving it from several initializations gives a candidate set, and a prediction-conditioned oriented-bounding-box (OBB) feasibility test retains only candidates whose minimum predicted OBB clearance over the horizon meets a threshold. The lower layer executes the lowest-cost survivor, or a route-centerline backup when none remains, through the tracking controller supplied with the planner, augmented by a range-based speed bound and a saturated proportional braking law. In closed-loop simulation on 126 Bench2Drive routes with VAD as the upstream planner, CLRE raises the driving score from 43.41 to 56.42 and route completion from 57.27 to 72.23, and reduces collision events from 70 to 53.
Figures & tables
Fig. 1: Local coordinate frame, ego state, and control variables used for trajectory refinement. The frame is fixed at the ego pose at replanning instant t . The positive position axes point rightward and forward, and positive curvature corresponds to a right turn.
Fig. 2: Overview of CLRE. Gray blocks are supplied unchanged with the upstream planner, colored blocks are introduced by CLRE, and dark blocks denote sensor inputs and the vehicle.
Fig. 3: Outer and inner interaction regions around a predicted agent.
Fig. 4: Effect of refinement on Route 16390. (a) Executing the nominal VAD trajectory directly, the vehicle remains at its spawn point with a zero-length plan and full brake applied at t=6.95 s. (b) At the same instant, CLRE is traveling at 6.1 m/s. (c) Executed paths over the whole route: CLRE completes it, whereas the baseline covers no distance and stalls until timeout.
Fig. 5: Prediction-conditioned OBB feasibility test on Route 17749, shown with the predicted agent, pedestrian, and ego boxes at k=0 and at 3 s ( k=N ). (a) The lowest-cost candidate overlaps the predicted pedestrian box and is rejected, with gap=−0.30 m. (b) The lowest-cost feasible candidate is selected, with gap=+0.96 m.
Method
Routes
DS ↑
RC ↑
SR ↑
Coll. ↓
VAD [ 2 ]
126
43.41
57.27
22.22%
70
CLRE-VAD
126
56.42
72.23
28.57%
53
UniAD [ 1 ]
11
30.34
58.97
0.00%
17
CLRE-UniAD
11
40.81
61.00
9.09%
9
DriveTransformer [ 23 ]
11
39.20
85.40
9.09%
20
CLRE-DriveTransformer
11
52.30
71.14
9.09%
5
TABLE I: Overall closed-loop performance.
Variant
DS
RC
Compl.
Coll.
Coll./km
Brake
Jerk
↑
↑
↑
↓
↓
↓
↓
Planner ablation
Full
63.57
75.22
22
10
3.18
0.042
28.7
w/o progress
53.13
57.59
15
2
0.79
–
–
w/o feas. test
58.30
74.39
20
17
5.40
–
–
Execution-layer ablation
TABLE II: Ablations of CLRE-VAD on the first 36 routes.
Fig. 6: Winner retention under single-weight scaling over 41,858 development solves with at least two distinct feasible candidates. Retention is the fraction of solves whose selected candidate is unchanged from the nominal setting α=1 .
Candidate
Feasibility
Dev. at
Max dev.
Comparison
change
agreement
k=2 (m)
(m)
2× iteration budget
51.8%
99.97%
0.005
0.11
5× iteration budget
58.7%
99.91%
0.006
0.13
TABLE III: Solver sensitivity over 3275 replanning windows.
Department of Automation, University of Science and Technology of China · Institute for AI Industry Research, Tsinghua University · School of Electronic Information Engineering, Beihang University +1
Institute of Measurement and Control Systems, Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany · FZI Research Center for Information Technology, Karlsruhe, Germany