RUL-Aware RRT*: Degradation-Balanced Motion Planning for Robotic Manipulators

1 Introduction

• Uneven motor use can accelerate degradation in heavily worked joints, making one motor a system bottleneck whose failure shuts down the robot and disrupts production.The introduction links this risk to planning optimized for geometry or energy without accounting for joint health.Source: §1 · p. 1
• The introduction identifies three gaps: health-blind planning, predictive maintenance largely separated from motion planning, and limited attention to uneven multi-motor degradation and lifetime bottlenecks.Existing planning studies are described as focusing mainly on energy efficiency or collision avoidance.Source: §1 · p. 2
• RUL-aware RRT* adds an RUL-aware cost function, online joint-RUL updates, and adaptive joint weighting to the standard RRT* framework.The weighting mechanism is intended to balance motor loads under different degradation conditions.Source: §1 · p. 2
• Across three UR5 scenarios, RUL-aware RRT* balances degradation, enhances health awareness, and extends the first-replacement time of critical motors.The scenarios are Full Health, Heterogeneous Degradation, and Local Degradation.Source: §1 · p. 2

2 Background and Preliminaries

• RRT* incrementally samples and extends a tree, then rewires connections to refine a collision-free path toward the goal while retaining probabilistic completeness and asymptotic optimality.The paper describes it as a prominent framework for feasible, near-optimal trajectories in high-dimensional configuration spaces.Source: §2 · p. 3
• Expansion samples a free configuration, selects the nearest tree node by connection cost, and steers from it to create a new node.The new node is then considered for connection to nearby nodes.Source: §2 · p. 3
• Parent selection chooses the neighboring node with the lowest accumulated path cost to the new node, subject to collision-free connection.Accumulated cost sums incremental costs along the tree path from the start.Source: §2 · pp. 3–5
• Rewiring reconnects a nearby node through the new node when that route lowers its total cost, incrementally refining the tree.In standard RRT*, the local and accumulated costs are based on path length, measured as Euclidean distance in joint space.Source: §2 · p. 4

3.1 Overview of the RUL-aware Motion Planning Framework

• The framework connects planning, execution, and health monitoring: it plans a collision-free path using joint RUL, then uses post-execution feedback to adapt later plans.RUL and adaptive weights enter before planning, during cost evaluation, and after execution through feedback.Source: §3.1 · pp. 5–6

3.2 RUL-aware Cost Formulation

• The RUL-aware cost combines joint-space distance with a health penalty, discouraging motion on degraded joints while balancing path efficiency and reliability.The penalty increases for joints with shorter RUL, and the formulation is used to evaluate candidate paths.Source: §3.2 · pp. 6–7

3.3 Online Joint RUL Update Mechanism

• After each planning–execution cycle, operational and health feedback updates joint RUL, supplying current health information for the next planning iteration.The general update formulation supports different strategies, including predictive and hybrid recursive updates.Source: §3.3 · pp. 7–8

3.4 RUL-driven Adaptive Joint Weighting Mechanism

• Joint weights increase for lower-RUL actuators, while normalization keeps the average weight at one and imbalance controls how sharply weak joints are emphasized.When joint RUL values are similar, weights are nearly uniform; greater imbalance produces a more peaked allocation.Source: §3.4 · pp. 8–9

3.5 Integration with RRT

• RUL-aware costs modify expansion, parent selection, and rewiring, while feedback returns updated RUL and weights to subsequent planning cycles.These changes retain RRT*’s sampling-based structure while evaluating tree connections for both geometric efficiency and joint health.Source: §3.5 · pp. 9–10

4.1 Experimental Setup

• The planner is evaluated in ROS Melodic with MoveIt and OMPL on a simulated 6-DOF UR5; joint angular displacement serves as a usage-based RUL feedback surrogate.The closed loop plans and executes pick-and-place motions, then updates RUL and joint weights; real sensor measurements are not collected.Source: §4.1 · pp. 10–11

4.2 Test Scenarios

• The scenarios initialize all joints at RUL 1000 in FHS, [500, 800, 400, 600, 1000, 700] in HDS, and [100, 1000, 1000, 1000, 1000, 1000] in LDS.Degradation curvature varies with p = 0.8, 1, or 1.5; experiments run for 80 tasks or until a joint reaches RUL 0, with LDS run to weakest-joint failure.Source: §4.2 · p. 12

4.3 Evaluation Metrics

• Evaluation metrics cover total RUL consumption, remaining RUL, degradation CV, weakest-joint lifetime, lifetime recovery, first-failure task index, relative lifetime gain, and balance at failure.Lower total consumption and CV indicate efficiency and more synchronized degradation; higher remaining or bottleneck RUL and a larger first-failure index indicate improved lifetime or delayed failure.Source: §4.3 · pp. 13–14

(1) Full Health Scenario

• In FHS at p = 0.8, RULmin rises from approximately 704 with Baseline RRT* to approximately 733 with RUL-aware RRT*, a Grec of approximately 4.1%.Recovery ratios at p = 1.0 and p = 1.5 are approximately 1.7% and 0.5%, respectively.Source: (1) Full Health Scenario · p. 14, Figure 5 · p. 17

(2) Heterogeneous Degradation Scenario

• In HDS at p = 0.8, RULmin rises from approximately 164 with Baseline RRT* to approximately 203 with RUL-aware RRT*, a Grec of approximately 23.8%.Recovery ratios at p = 1.0 and p = 1.5 are approximately 25.7% and 29.2%, respectively.Source: (2) Heterogeneous Degradation Scenario · p. 16, Figure 5 · p. 17

(1) Full Health Scenario

• In FHS, Baseline RRT* CVRUL reaches 0.03–0.06, while RUL-aware CVRUL stays near zero; remaining RUL sum improves by approximately 100, 238, and 287 units for p = 0.8, 1.0, and 1.5.The RUL-aware planner typically increases total RUL consumption by about 100–200 units after 80 tasks, while producing a healthier final state.Source: (1) Full Health Scenario · p. 16, Figure 7 · p. 19, Figure 8 · p. 19

(2) Heterogeneous Degradation Scenario

• In HDS, health-aware weighting shifts effort toward healthier joints, while CVRUL stays within 0.25–0.35 versus 0.40–0.55 for Baseline RRT*.The redistribution slows degradation of weaker components and prevents initial health differences from amplifying as strongly.Source: Figure 9 · p. 20, (2) Heterogeneous Degradation Scenario · p. 20
• Across HDS and LDS, continuous health feedback reallocates motion away from weak joints, increasing their weights as health declines and protecting bottleneck lifetime by slowing degradation.In HDS, this counteracts the tendency for degradation to concentrate on initially weak joints; in LDS, it reduces the weakened joint’s involvement in subsequent trajectories.Source: (2) Heterogeneous Degradation Scenario · p. 20, (3) Local Degradation Scenario · pp. 21–22
• In HDS, remaining lifetime sum RUL(T) rises approximately 185, 273, and 302 units at p = 0.8, 1.0, and 1.5, despite only 80–150 extra RULsum consumption.The passage describes this as a favorable trade-off: modest additional system-wide lifetime use yields a higher lifetime margin for the bottleneck component.Source: (2) Heterogeneous Degradation Scenario · p. 20, Figure 10 · p. 21

(3) Local Degradation Scenario

• In LDS, RUL-aware RRT* completes 27.7 vs 18.7 tasks (+48.1%) at p = 0.8, 51.7 vs 33.0 (+56.7%) at p = 1.0, and 111.7 vs 73.7 (+51.6%) at p = 1.5.These gains correspond to a consistently higher first failure task index kfail across all three degradation parameters.Source: (3) Local Degradation Scenario · p. 21, Figure 11 · p. 22
• Across scenarios, RUL-aware planning yields larger gains as degradation becomes more nonlinear, while still providing noticeable benefits under milder wear dynamics.The passages attribute this pattern to health-based reweighting being more valuable when additional use of an already weak joint causes disproportionately large RUL loss.Source: (3) Local Degradation Scenario · pp. 22–23
• In LDS, CVRUL remains around 0.40–0.42 for RUL-aware RRT* versus 0.44–0.46 for Baseline RRT* near failure, and failure-time joint-RUL spread is lower under the proposed planner.A smaller failure-time coefficient of variation indicates a more balanced degradation state at system failure.Source: (3) Local Degradation Scenario · pp. 21–22, Figure 12 · p. 23

Full Health Scenario

• In FHS, RUL-aware RRT* maintains balanced degradation, raises minimum remaining lifetime RULmin, reduces joint-RUL variation, and improves Grec with a small increase in total lifetime consumption.Source: §5 · p. 23

Heterogeneous Degradation Scenario

• In HDS, the planner protects initially weak joints, limits health disparity amplification, and yields higher bottleneck lifetimes, more uniform degradation, and lower CVRUL.The passage reports a healthier system state at the end of execution.Source: §5 · p. 23

Local Degradation Scenario

• In LDS, the first failure task index kfail rises by 48–56% across degradation parameters, and degradation balance at failure also improves.The passage also reports meaningful relative lifetime gains and smaller CV_fail values.Source: §5 · p. 23

Degradation sensitivity

• Reliability gains are larger when degradation is more sensitive to accumulated motion usage, while benefits remain noticeable in milder regimes.Source: §5 · p. 24

Conclusion

• Overall, the method balances degradation, delays bottleneck formation, and improves lifetime performance under wear.Source: §5 · p. 24

Future Work

• The authors propose extending RUL-aware RRT* to multi-robot shared tasks, with agentic AI and reinforcement learning for adaptive motion allocation and coordinated health-aware decisions.Source: §5 · p. 24

• The authors used ChatGPT 4o to partially refine the completed manuscript.Source: §6 · p. 24
• The authors reviewed and edited the manuscript after using the tool.Source: §6 · p. 24
• The authors state that they take full responsibility for the published article’s content.Source: §6 · p. 24