Rehabilitation robot trajectory planning method for upper limb based on healthy limb motion using multi-objective constrained reinforcement learning | International Journal for Simulation and Multidisciplinary Design Optimization (IJSMDO)

Open Access

Table 4

Ablation study of reward function components and final weight configuration.

Stage / Reward component added	Final Avg. reward	SD of Jerk (rad/s³) (Joint 3 avg.)¹	Tracking error (rad) (Avg. RMSE)
Baseline (Tracking only)	15.2	45.8	0.008
+ Smoothness & Jerk	18.5	19.5	0.012
+ Velocity Barrier	19.1	18.9	0.011
Full Model (+ Keypoints, etc.)	19.8	18.2	0.007

Final weights used: w_track=2.0, w_smooth=1.0, w_jerk=1.0, w_vel=1.5, w_manip=1.2, w_soft=0.8, w_hard=2.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.