---
title: 'BOWConnect: Bidirectional Kinodynamic Planner'
url: https://www.emergentmind.com/topics/bowconnect
type: topic
---

# BOWConnect: Bidirectional Kinodynamic Planner

Searching arXiv for the BOWConnect paper and closely related kinodynamic motion-planning work.

BOWConnect is a **bidirectional, parallel kinodynamic motion planner** that integrates **Bayesian Optimization over Windows (BOW)** as a learning-based steering function within a tree-based sampling architecture for generating **collision-free, dynamically feasible trajectories** in continuous state and control spaces [2606.27292]. It is designed for planning problems in which motion must satisfy differential constraints, so the relevant search domain is the full state space \(\mathbf{x}=(q,\dot q)\in\mathcal{X}\), not merely configuration space. The method is explicitly motivated by three limitations of existing sampling-based kinodynamic planners: **sample inefficiency in high-dimensional state spaces**, **unreliable cost heuristics under dynamic constraints**, and **poor performance in narrow passage environments** [2606.27292]. In the formulation reported for BOWConnect, multiple forward and backward workers grow trees in parallel from sampled start and goal-region states, while a spatial hashing scheme accelerates connection discovery and a boundary value problem solver produces kinodynamically consistent bridge trajectories [2606.27292].

## 1. Planning setting and design motivation

BOWConnect addresses the kinodynamic motion-planning problem in which trajectories must remain inside the collision-free feasible state space while satisfying system dynamics of the form
\[
\dot{\mathbf{x}} = f(\mathbf{x}, \mathbf{u}),
\]
with admissible controls \(\mathbf{u}\in\mathcal{U}\subset\mathbb{R}^m\) [2606.27292]. The paper defines the configuration space \(\mathcal{C}\), the collision-free subset \(\mathcal{C}_{\text{free}}\), the state space \(\mathcal{X}=T\mathcal{C}\), and the feasible state space
\[
\mathcal{X}_{\text{free}} = \{(q,\dot{q}) \in \mathcal{X} \mid q \in \mathcal{C}_{\text{free}}\}.
\]
A trajectory over horizon \(T\) induced by a control \(\mathbf{u}\) is denoted
\[
\tau(\mathbf{u}) = \Phi(\mathbf{x}, \mathbf{u}) = \{\mathbf{x}(t) : t \in [0, T]\},
\]
and the planning objective is to find a feasible \(\tau(\mathbf{u})\subset\mathcal{X}_{\text{free}}\) that reaches the goal state [2606.27292].

The planner is positioned against classical kinodynamic methods such as RRT, SST, EST, and KPIECE, as well as against the authors’ earlier unidirectional BOW method [2606.27292]. The central criticism is that many classical planners rely on **random control sampling** and weak geometric heuristics, whereas geometric proximity is often a poor surrogate for actual transition cost or reachability under nonholonomic and dynamic constraints [2606.27292]. The paper therefore replaces random local steering with an online learning procedure over short control horizons and embeds it in a global bidirectional exploration framework.

A plausible implication is that BOWConnect is best understood not as a purely local optimizer or a purely global sampling planner, but as a hybrid architecture in which local model-based learning and global tree connectivity are coupled at every stage. The paper’s direct comparison to the earlier BOW planner suggests that the addition of bidirectionality, worker-level parallelism, and explicit connection machinery is intended to remedy the global exploration limitations of a unidirectional windowed Bayesian-optimization planner [2606.27292].

## 2. Bayesian Optimization over Windows as local steering

In BOWConnect, **BOW** denotes the use of **Bayesian optimization over short finite-horizon control windows** [2606.27292]. Rather than optimizing an entire long trajectory globally, the planner optimizes over a local horizon \(T\), starting from a current state \(\mathbf{x}_{\text{near}}\), over controls \(\mathbf{u}\in\mathcal{U}\), using forward simulation under the dynamics and online surrogate modeling in the control space [2606.27292]. This local optimizer serves as the planner’s steering function.

The paper defines a local reward
\[
r(\mathbf{u}) =
\begin{cases}
-\|\mathbf{p}(T)-\mathbf{p}_{\text{target}}\| & \text{if trajectory is collision-free} \\
-\infty & \text{otherwise}
\end{cases}
\tag{1}
\]
and a feasibility indicator
\[
c(\mathbf{u}) = 1 - \mathbb{I}_{\text{collision}(\tau(\mathbf{u}))}.
\tag{2}
\]
Two Gaussian process surrogates are then learned: \(\mathcal{GP}_r\) for reward and \(\mathcal{GP}_c\) for feasibility, with predictive moments
\[
\mu_r(\mathbf{u}),\ \sigma_r^2(\mathbf{u}), \qquad \mu_c(\mathbf{u}),\ \sigma_c^2(\mathbf{u}).
\]
The estimated probability of feasibility is
\[
P_{\text{feas}}(\mathbf{u}) = \Phi\left(\frac{\mu_c(\mathbf{u})}{\sigma_c(\mathbf{u})}\right),
\tag{3}
\]
and the constrained acquisition function used to select the next control is
\[
\alpha(\mathbf{u}) = \big[\mu_r(\mathbf{u}) - \kappa \sigma_r(\mathbf{u})\big] \times P_{\text{feas}}(\mathbf{u}).
\tag{4}
\]
The control maximizing \(\alpha(\mathbf{u})\) is used to generate the next short-horizon motion primitive [2606.27292].

The paper describes the resulting surrogate models as **local cost maps and constraints**, but these are not global workspace maps. They are local probabilistic surrogates over controls, encoding which controls are likely to make progress toward the target and which are likely to remain feasible in the current local planning context [2606.27292]. This distinction is important: the learned object is a local model of the short-horizon reachable structure, not a persistent global value function.

This suggests that the BOW component is effectively a dynamics-aware receding-horizon local planner embedded inside a sampling framework. The article’s terminology of “windows” therefore refers to short planning horizons in time and control space, not to fixed geometric windows in configuration space [2606.27292].

## 3. Bidirectional parallel architecture

BOWConnect is not a single-tree planner. It uses a **bidirectional architecture** with multiple workers operating in parallel, divided into \(N/2\) forward workers and \(N/2\) backward workers [2606.27292]. The planner first samples multiple start-region states and goal-region states:
\[
\mathcal{S} \leftarrow \text{SampleStates}(\mathbf{x}_{\text{start}}, r_{\text{goal}}, N),
\]
\[
\mathcal{G} \leftarrow \text{SampleStates}(\mathbf{p}_{\text{goal}}, r_{\text{goal}}, N).
\]
These states are collision-free samples within a radius around the start and goal regions, each with a random heading \(\theta \in [-\pi,\pi]\), which the paper states is important for resolving heading ambiguity at the goal and diversifying exploration [2606.27292].

Each worker maintains its own search tree and its own BOW instance [2606.27292]. Repeatedly, a worker samples a random state \(\mathbf{x}_{\text{rand}}\), finds the nearest node \(\mathbf{x}_{\text{near}}\) in its tree, invokes BOW to generate a short-horizon collision-free, dynamically feasible trajectory from \(\mathbf{x}_{\text{near}}\) toward \(\mathbf{x}_{\text{rand}}\), and appends all states along that local trajectory if the attempt succeeds [2606.27292].

The main thread performs inter-tree connection checks while workers expand independently. Nearby states from forward and backward trees are detected using spatial hashing, filtered using a kinematic feasibility test, and then passed to a boundary value problem solver to generate a bridge trajectory [2606.27292]. If a bridge succeeds, the final solution is assembled by merging the forward path, bridge, and backward path. The planner also includes a unidirectional fallback: if a single worker reaches the goal region directly, that solution may be returned [2606.27292].

A heuristic explanation for the benefit of many forward/backward workers is given by the idealized expression
\[
P_{\text{success}} = 1 - (1 - p_{\text{single}})^{N_f \times N_b},
\tag{9}
\]
where \(p_{\text{single}}\) is the success probability of a single forward/backward worker pair [2606.27292]. The paper does not present this as a completeness theorem, but as an explanation of a portfolio effect due to parallel diversity.

## 4. Connection machinery: spatial hashing and bridge generation

A central architectural element is the use of **spatial hashing** to accelerate candidate connection discovery between forward and backward trees [2606.27292]. Each state is mapped to a discretized grid cell:
\[
h(\mathbf{x}) = \left\lfloor \frac{x-x_0}{\Delta} \right\rfloor \oplus
\left\lfloor \frac{y-y_0}{\Delta} \right\rfloor \oplus
\left\lfloor \frac{z-z_0}{\Delta} \right\rfloor,
\tag{5}
\]
where \(\Delta\) is grid resolution, \((x_0,y_0,z_0)\) is a reference offset, and \(\oplus\) denotes bit-packing operations. The paper states that this yields \(\mathcal{O}(1)\) average-case lookup for nearby candidate states [2606.27292].

Before invoking a bridge solver, the planner checks a kinematic feasibility condition between a forward-tree state \(\mathbf{x}_f\) and a backward-tree state \(\mathbf{x}_b\):
\[
\left|\theta_{\text{req}} - \theta_f\right| < \theta_{\max}
\quad \text{and} \quad
\frac{\|\mathbf{p}_b - \mathbf{p}_f\|}{v_{\max}} < T_{\max},
\tag{6}
\]
with
\[
\theta_{\text{req}} = \arctan2(y_b-y_f,\, x_b-x_f).
\]
Only if this coarse test succeeds does the planner attempt to solve a local connection problem [2606.27292].

The bridge trajectory is generated by a boundary value problem solver seeking a control \(\mathbf{u}_c:[0,t_c]\to\mathcal{U}\) such that
\[
\mathbf{x}(t_c) = \mathbf{x}_b, \quad \tau(\mathbf{u}_c) \subset \mathcal{X}_{\text{free}}.
\tag{7}
\]
The reported solver is not an exact analytic steering law. Instead, it uses proportional heading control with heading error
\[
\Delta\theta = \theta_{\text{desired}} - \theta_{\text{current}},
\]
and yaw-rate command
\[
\omega = \text{clamp}\left(\frac{k_p \Delta\theta}{\Delta t},
-\omega_{\max}, \omega_{\max}\right).
\tag{8}
\]
Forward speed is set proportional to remaining distance, and the resulting bridge is collision checked before acceptance [2606.27292].

This structure makes BOWConnect distinct from purely geometric bidirectional planners. The connection phase is not based on straight-line interpolation between states, but on a dynamically consistent local bridge, with hashing and kinematic filtering used to make repeated connection attempts tractable [2606.27292].

## 5. State, control, and system models

The general state representation used by the planner is
\[
\mathbf{x}=(q,\dot q)\in\mathcal{X},
\]
with robot-specific instantiations given for both ground and aerial vehicles [2606.27292].

For the physical unicycle ground-vehicle experiments, the state is
\[
\mathbf{x}=(x,y,\theta,v,\omega),
\]
with control
\[
\mathbf{u}=(v_c,\omega_c).
\]
For the physical quadrotor experiments, the state is
\[
\mathbf{x}=(x,y,z,\theta,\dot x,\dot y,\dot z,\dot\theta),
\]
with control
\[
\mathbf{u}=(\dot x_c,\dot y_c,\dot z_c,\dot\theta_c).
\]
The paper states that fourth-order Runge–Kutta integration is used for forward simulation under the dynamics [2606.27292].

Across the benchmark evaluations, the reported planner parameters include speed limits from \(0.0\) to \(1.0\) m/s, maximum acceleration \(0.5\) m/s\(^2\), maximum yaw rate \(0.6981\) rad/s, maximum yaw acceleration \(2.0472\) rad/s\(^2\), time discretization \(0.1\) s, and local planning horizon \(3.0\)–\(5.0\) s [2606.27292]. The planner inputs also include the number of workers \(N\), a goal sampling radius \(r_{\text{goal}}\), and total time budget \(T_{\max}\) [2606.27292].

The global problem statement allows a time-optimal interpretation in which \(t_F\) is minimized over feasible solutions, but the paper does not claim asymptotic optimality for BOWConnect [2606.27292]. Its stated emphasis is fast feasible planning with high success and favorable trajectory quality.

## 6. Experimental evaluation

The planner is evaluated on **ten benchmark environments**: six UGV environments—Bugtrap, Narrow Passage-1, Narrow Passage-2, Forest, Nonconvex, and Intel—and four UAV environments—Quad-1, Quad-2, Quad-3, and Quad-4 [2606.27292]. Baselines are RRT, SST, EST, KPIECE, and BOW [2606.27292]. Metrics include planning time, trajectory length, success rate, average velocity, and average jerk [2606.27292]. UGV planners are tested under both **Unicycle** and **Bicycle** motion models [2606.27292]. The reported hardware platform is Ubuntu 22.04 on an AMD Ryzen 9 7950X with 32 GB RAM, and five trials are run per environment-planner pair [2606.27292].

The paper states that **BOWConnect achieves 100\% success in all six UGV environments under both Unicycle and Bicycle models** [2606.27292]. In Bugtrap, the reported planning times are \(0.035 \pm 0.014\) s for Unicycle and \(0.021 \pm 0.005\) s for Bicycle, both with 100% success [2606.27292]. In Narrow Passage-1, BOWConnect achieves \(0.048 \pm 0.003\) s for Unicycle and \(0.034 \pm 0.008\) s for Bicycle, both with 100% success, whereas the paper reports strong degradation of several baselines under Bicycle dynamics, including 0% success for EST, KPIECE, and BOW [2606.27292]. In Narrow Passage-2, the corresponding BOWConnect times are \(0.056 \pm 0.021\) s and \(0.048 \pm 0.011\) s, again both with 100% success [2606.27292].

In Forest, BOWConnect achieves \(0.024 \pm 0.001\) s for Unicycle and \(0.025 \pm 0.001\) s for Bicycle, both with 100% success, and the shortest average trajectories among reported methods: \(20.176\) m for Unicycle and \(21.464\) m for Bicycle [2606.27292]. In Nonconvex, it achieves \(0.024 \pm 0.007\) s for Unicycle and \(0.019 \pm 0.004\) s for Bicycle, again with 100% success, and shortest average trajectories of \(15.814\) m and \(17.816\) m, respectively [2606.27292]. The paper explicitly attributes these shorter paths in Nonconvex to backward-tree guidance [2606.27292]. In the Intel benchmark, described as the most complex, BOWConnect achieves \(0.397 \pm 0.039\) s for Unicycle and \(2.133 \pm 1.707\) s for Bicycle, with 100% success in both cases [2606.27292].

For UAV environments, the paper states that all planners achieve 100% success across Quad-1 to Quad-4, but BOWConnect and BOW are much faster than the classical baselines [2606.27292]. In Quad-1, BOWConnect achieves \(0.117 \pm 0.092\) s with trajectory length \(6.562\) m and lowest jerk \(0.0279\) [2606.27292]. In Quad-2, it is the fastest method at \(0.086 \pm 0.069\) s [2606.27292]. In Quad-3, it achieves \(0.081 \pm 0.029\) s [2606.27292]. In Quad-4, it achieves \(0.091 \pm 0.078\) s, trajectory length \(6.231\) m, and lowest jerk \(0.0246\) [2606.27292].

The paper does not include a dedicated ablation table isolating individual components such as spatial hashing or bidirectionality [2606.27292]. The closest comparison is against the earlier unidirectional BOW planner, which the paper uses to show the effect of adding the bidirectional parallel architecture and connection machinery [2606.27292].

## 7. Real-world deployment

The reported real-world ground platform is an **iRobot Create 3 differential-drive educational robot** with radius \(0.17\) m operating in a \(6.5\text{ m}\times 5.5\text{ m}\) cluttered indoor laboratory [2606.27292]. Localization is provided by Vicon motion capture at 120 Hz, and collision checking is performed against a 2D occupancy grid derived from the laboratory layout [2606.27292]. The planner uses the unicycle motion model, goals are sent interactively via RViz 2D goal commands, maximum linear velocity is \(1.0\) m/s, maximum angular velocity is \(1.5\) rad/s, and the low-level control step is \(\Delta t = 0.05\) s [2606.27292]. The paper reports that feasible trajectories are consistently generated in under \(0.15\) s, with average planning time \(0.12\) s, maximum \(0.15\) s, and **no collisions or tracking instabilities** across five independent trials [2606.27292].

The reported aerial platform is a **Parrot Bebop 2 UAV** operating in a \(6.5\text{ m}\times 5.5\text{ m}\times 2.5\text{ m}\) indoor environment containing three box-shaped obstacles: two of size \(0.4\times 0.8\times 0.92\) m and one of size \(0.64\times 0.64\times 1.08\) m [2606.27292]. Goals are sent from RViz using the Nav Goal tool [2606.27292]. The paper states that planned and executed trajectories align well and that collision-free trajectories are consistently computed in under \(0.1\) s [2606.27292].

These deployments are used as evidence that the planner can operate in real time on both a ground vehicle and a quadrotor under the state and control models described above [2606.27292].

## 8. Interpretation, significance, and limitations

BOWConnect’s central methodological claim is that **learned local cost and feasibility models improve sample efficiency**, while **parallel bidirectional exploration improves global connectivity**, especially in narrow passages and non-convex spaces [2606.27292]. The acquisition function
\[
\alpha(\mathbf{u})= [\mu_r(\mathbf{u})-\kappa\sigma_r(\mathbf{u})]P_{\text{feas}}(\mathbf{u})
\]
is the mechanism by which the local planner balances progress toward the target against estimated collision risk [2606.27292]. The paper’s explanation of narrow-passage behavior is that repeated collisions reduce the feasibility-weighted score of controls heading directly into obstacles, causing the optimizer to shift toward controls that remain inside free space [2606.27292].

The planner is also presented as a practical way to make bidirectional kinodynamic planning effective despite the absence of an exact two-point steering function [2606.27292]. In this respect, the bridge solver and hashing-based connection layer are as central to the method as the local Bayesian optimization component.

Several limitations are either stated or implied. The bridge solver is heuristic rather than exact, relying on proportional control rather than an analytic kinodynamic steering solution [2606.27292]. Actual parallel speedup is limited by connection checking overhead and sequential BVP solving [2606.27292]. The current scope is **static environments**, with dynamic obstacles identified as future work, and planning under uncertainty is likewise listed as future work [2606.27292]. The paper does not provide asymptotic optimality or strong formal completeness guarantees specific to BOWConnect [2606.27292]. It also omits several lower-level implementation details, including kernel choice, GP hyperparameter fitting procedure, number of BO iterations per window, exact acquisition optimizer, exact worker count used in benchmarks, and exact hash resolution \(\Delta\) [2606.27292].

A common misconception would be to interpret BOWConnect as merely a parallelized version of BOW or a kinodynamic analogue of RRT-Connect. The paper argues for a more specific characterization: BOWConnect is a planner in which **online constrained Bayesian optimization** over short windows is the local steering primitive, **bidirectional parallel search** supplies global exploration and connection opportunities, **spatial hashing** supplies efficient cross-tree candidate discovery, and a **BVP solver** enforces kinodynamic consistency at the connection stage [2606.27292].

## 9. Position within related motion-planning work

The paper situates BOWConnect relative to classical kinodynamic planners such as RRT, EST, KPIECE, and SST, which it characterizes as being heavily dependent on random control sampling and weak geometric heuristics [2606.27292]. It also positions itself relative to prior BO-based or GP-based planners, describing them as data efficient but typically computationally expensive and often limited to unidirectional or locally greedy use [2606.27292]. Compared with the earlier **BOW** planner, BOWConnect adds bidirectional search, multiple forward and backward workers, tree-based exploration, spatial-hash-assisted connection discovery, and BVP bridge creation [2606.27292].

This suggests that BOWConnect should be regarded as an integration of several distinct strands of planning methodology: model-based local optimization over controls, tree-based kinodynamic exploration, multithreaded search diversification, and approximate boundary-value connection methods. The closest antecedent identified by the paper is its own earlier unidirectional BOW framework, but the architectural scope of BOWConnect is substantially broader [2606.27292].

The paper states that videos of real-world and simulated experiments, high-resolution figures, and open-source code are available at the project site, indicating an emphasis on practical reproducibility as well as algorithmic novelty [2606.27292].

Source: https://www.emergentmind.com/topics/bowconnect