Non-prehensile transportation by robotic arms is a promising approach for the service industry. Such applications require robotic arms to be reactive, safe, and flexible enough to navigate dynamically changing and complex environments while maintaining the transported objects' stability. However, existing limited-horizon motion control methods are often ineffective in these scenarios. This letter presents a risk-adaptive motion planning framework for safe and resilient non-prehensile object transportation. To adapt to disturbances in quasi-static environments, the framework integrates a three-level safety mechanism that responds to predicted time-to-collision with replanning, a planned graceful safety stop, or a hardware-level emergency stop. The former two maneuvers rely on an efficient trajectory generator that integrates constrained path planning, trajectory parameterization, and refinement. Specifically, the constrained path planner is employed to provide global guidance in complex environments, yielding a high-quality initialization for the subsequent coarse-to-fine trajectory generation strategy. Additionally, the framework incorporates a periodic replanning strategy to proactively adapt to dynamic goal switching and environmental changes. Finally, the framework is comprehensively evaluated through extensive experiments.
Risk-adaptive motion planning framework, featuring a dual-triggering mechanism (periodic and safety-driven) and a complete trajectory generation pipeline, which integrates constrained path planning, trajectory parameterization, and trajectory refinement.
Note: All videos are shown at real-time speed (1x) unless otherwise stated.
We compare our method (left) with a strong reactive baseline (right) adapted from an MPC approach [1] in four scenarios.
Ours (An open space)
Baseline (An open space)
Ours (A narrow passage)
Baseline (A narrow passage)
Ours (A table obstacle)
Baseline (A table obstacle)
Ours (A cluttered scene)
Baseline (A cluttered scene)
The performance of the replanning framework is evaluated in dynamic environments.
The replanning maneuver in response to a sudden obstacle
The safety-stop maneuver in response to a sudden obstacle
Non‑prehensile transportation in a dynamic environment (scene updated every 3s)
We evaluate the robustness of our method against CoM displacement under varying levels of motion aggressiveness.
Robustness evaluation against position deviations, tested at three aggressiveness levels (low, medium, and high). (a-c) Heatmaps show the deviation between the actual and nominal contact force angles, resulting from the applied CoM position deviations. (d) The experimental setup. (e) Corresponding success rates.
Low aggressiveness,
Deviation: (x = 0.0 m, y = 0.0 m)
Medium aggressiveness,
Deviation: (x = 0.0 m, y = 0.0 m)
High aggressiveness,
Deviation: (x = 0.0 m, y = 0.0 m)
Low aggressiveness,
Deviation: (x = 0.05 m, y = 0.0 m)
Medium aggressiveness,
Deviation: (x = 0.05 m, y = 0.0 m)
High aggressiveness,
Deviation: (x = 0.05 m, y = 0.0 m)
Low aggressiveness,
Deviation: (x = 0.1 m, y = 0.0 m)
Medium aggressiveness,
Deviation: (x = 0.1 m, y = 0.0 m)
High aggressiveness,
Deviation: (x = 0.1 m, y = 0.0 m)
Low aggressiveness,
Deviation: (x = 0.13 m, y = 0.0 m)
Medium aggressiveness,
Deviation: (x = 0.13 m, y = 0.0 m)
High aggressiveness,
Deviation: (x = 0.13 m, y = 0.0 m)
For the cuboid (5 × 5 × 20 cm) with a tipping margin angle β ≈ 14.0°, the constraint violation is insufficient to induce tipping. However, for a slender cuboid 2.5 × 2.5 × 30 cm, β ≈ 4.8°, the risk of tipping increases significantly.
Finally, the proposed framework is evaluated in a series of challenging real-world scenarios. First, we examine the performance across various tipping margin angles and motion aggressiveness levels. The following video is shown at 4x speed.
Low aggressiveness,
Transported Object (2.5 × 2.5 × 15 cm)
Medium aggressiveness,
Transported Object (2.5 × 2.5 × 15 cm)
High aggressiveness,
Transported Object (2.5 × 2.5 × 15 cm)
Low aggressiveness,
Transported Object (2.5 × 2.5 × 20 cm)
Medium aggressiveness,
Transported Object (2.5 × 2.5 × 20 cm)
High aggressiveness,
Transported Object (2.5 × 2.5 × 20 cm)
Low aggressiveness,
Transported Object (2.5 × 2.5 × 25 cm)
Medium aggressiveness,
Transported Object (2.5 × 2.5 × 25 cm)
High aggressiveness,
Transported Object (2.5 × 2.5 × 25 cm)
Low aggressiveness,
Transported Object (2.5 × 2.5 × 30 cm)
Medium aggressiveness,
Transported Object (2.5 × 2.5 × 30 cm)
High aggressiveness,
Transported Object (2.5 × 2.5 × 30 cm)
Success rates across various tipping margin angles and motion aggressiveness levels.
Subsequently, the reactivity and robustness of our method are evaluated by executing non-prehensile transportation of a slender bottle, milk cartons, and a partially filled water cup in a quasi-static environment. Notably, the implementation remains based on the cuboid model (5 × 5 × 20 cm) without object-specific parameter tuning.
Transport a slender bottle
Transport milk cartons
Transport a partially filled cup of water
Non-prehensile transportation with frequently changing goal points
To demonstrate the versatility and hardware-agnostic nature of our framework, we conducted further experiments on a Rokae XMate CR20 robotic arm. The results show that our method can be readily applied to different hardware with minimal adaptation.
Non-prehensile transportation in extremely cluttered environments
Transport two slender water bottles
Transport three slender water bottles
Transport two objects stacked vertically (front view)
Transport two objects stacked vertically (side view)
This letter presents a risk-adaptive framework for non-prehensile transportation in quasi-static environments subject to dynamic disturbances. To enhance reactivity and resilience, the framework integrates a three-level safety mechanism with proactive periodic replanning. The underlying planning pipeline employs a hierarchical coarse-to-fine architecture, where a constrained path planner provides global guidance and an unconstrained optimizer refines the trajectory, effectively balancing trajectory quality and planning efficiency. Extensive experiments demonstrate the framework's efficiency and robustness. Future work will aim to address the identified limitations and extend the approach to dynamic environments.