LEAN: Local Energy-Aware Navigation for Skid-Steer Robots Via Deep Reinforcement Learning
Aguila-Rigordi, O., Torres-Torriti, M., Toro, R., Cheein, F.A.A. and Araya, M. (2026) LEAN: Local Energy-Aware Navigation for Skid-Steer Robots Via Deep Reinforcement Learning. IEEE Robotics and Automation Letters. pp. 1-8. ISSN 2377-3774
|
Text
F Cheein LEAN local energy-aware navigation AAM OCR Upload.pdf - Accepted Version Download (1MB) | Preview |
Abstract
Energy-efficient local navigation of skid-steer mobile robots (SSMRs) in dense environments is constrained by a structural trade-off: the low-curvature, low-power maneuvers that minimize locomotion energy conflict with the high-curvature, high-yaw-rate corrections required to weave through dense obstacle fields without colliding. Model-based planners do not resolve this tension, but expose it as a cost knob, since weighting energy more heavily cuts consumption only by degrading performance as the scenes grow harder. We present the LEAN approach, a local planner based on deep reinforcement learning that learns locomotion-efficient navigation among obstacles from a calibrated energy signal while being model-free at inference, mapping observations directly to commands without solving any closed-form dynamic or power model online. The energy signal enters the policy only as raw motor currents, and in place of typical goal-progress dense reward shaping, the policy is guided by hindsight relabeling that keeps the reward pathshape- agnostic. Benchmarked using a 140-kg tracked SSMR against NMPC and EA-DWA variants that share the simulator's identified dynamics and power model, raising the energy weight of either baseline lowers consumption only by sacrificing success, leaving the most aggressive energy-aware baselines up to 14.2% success points below the proposed approach at the hard tier. LEAN instead reduces energy by 33.2±25.4% to 39.1±30.5% across training seeds in simulation and 11.9±4.1% in real-world experiments over the Vanilla-DRL (energy-blind) ablation at unchanged success rate, and selects actions faster than the modelbased controllers, with the simulation ranking preserved when transferred to cluttered outdoor environments. Compared to the energy-aware model-based NMPC-0.05, LEAN reduces energy by 6.3±4.9% in easy low-clutter environments up to 22.3±1.1% in harder high-clutter environments across training seeds in simulation and by 21.8±21.6% in real-world experiments.
| Item Type: | Article |
|---|---|
| Keywords: | Energy, Modeling, Navigation, Media, Training, Seeds (agriculture), Timing, Meters, Robots, Simulation |
| Divisions: | Departments > Engineering Research Centres > Harper Institute of Technology |
| Depositing User: | Mrs Susan Howe |
| Date Deposited: | 06 Nov 2026 11:05 |
| Last Modified: | 06 Nov 2026 11:05 |
| URI: | https://hau.repository.guildhe.ac.uk/id/eprint/18438 |
Actions (login required)
 |
Edit Item |