Chapter Four · failure evidence

What Trajectory & Motion Planning got wrong, from 79 dissertations

The records document widespread challenges encountered in trajectory and motion planning across diverse robotic platforms and numerical frameworks. Common failures stem from computational intractability in high-dimensional spaces, physical infeasibility caused by missing dynamic constraints, and algorithmic divergence during closed-loop execution. These records come from PhD theses at 17 institutions, 2021 to 2026. Each links to its thesis. They were extracted by language models reading the full text, so treat each as a lead to read, not a verdict.

Learning-based policies and neural network priors fail to generalize or maintain stable trajectory tracking

17 theses · 9 institutions

Pure neural network models, reinforcement learning policies, and imitation approaches often suffer from poor goal progress, high input dimensionality, and unaligned initial states. These methods frequently underperform classical controllers, cause training degradation across multi-waypoint tasks, or fail when executed directly on hardware without tracking controllers.

Tried and failed

zero-shot vision-language models applied to robot trajectory planning and motion reasoning. Outcome: did not generalise. Reason: models exhibit severe forward-motion and deceleration biases and fail at temporal motion reasoning

Building Intelligence that can Interact with the Physical World · MIT

Tried and failed

direct execution of diffusion-generated actions applied to robot trajectory control. Outcome: worse than baseline. Reason: generated action sequences execute poorly directly, requiring state-to-action tracking controllers to function reliably

Video as the Language of Embodied Intelligence · MIT

Tried and failed

behavior cloning from human demonstrations applied to autonomous ground vehicle goal navigation. Reason: learned obstacle avoidance behaviors but lacked sufficient goal-directed progress, causing consistent timeouts

Towards Improving and Extending Traditional Robot Autonomy with Human Guided Machine Learning · Virginia Tech

Tried and failed

supervised pretraining without reinforcement learning fine-tuning applied to robot trajectory generation and tracking. Outcome: worse than baseline. Reason: supervised learning alone fails to optimize multi-fidelity dynamics and trajectory tracking performance

Multi-fidelity Optimal Trajectory Generation: Optimal Experiment Design for Robot Learning · MIT

Tried and failed

test-time policy adaptation via action prediction applied to visual robot navigation. Outcome: worse than baseline. Reason: self-supervised action prediction auxiliary objective degraded navigation calibration performance during deployment

Harnessing Synthetic Data for Robust and Reliable Vision · Georgia Tech

Tried and failed

model-free reinforcement learning for continuous trajectory tracking applied to multirotor flight control. Outcome: worse than baseline. Reason: heading tracking accuracy and latency did not improve over waypoint-stabilization controllers

Deep Reinforcement Learning for Multirotor Flight Control: A Comparative Study of Sim-to-Real Training and Real-World Performance · Virginia Tech

Lost to a baseline

FCNet (pure neural network policy without BarrierNet guidance) achieved lower final robustness and objective value than BarrierNet, getting trapped in sub-optimal obstacle avoidance trajectories

Temporal logic robot control using machine learning · OpenBU

Lost to a baseline

PPO trajectory tracking policy did not provide substantive improvement in tracking performance compared to a standard nonlinear backstepping baseline.

Safe Approximately-Optimal High-Speed Autonomous Flight Control · JScholarship

Lost to a baseline

MBDRL was outperformed in navigation time by an aggressively tuned PID in single-drone navigation (though PID suffered from overshooting).

Cooperative Payload Transportation by UAVs: A Model-Based Deep Reinforcement Learning (MBDRL) Application · Virginia Tech

Considered and rejected

Considered and rejected: Training neural network car following models by predicting one step ahead from past empirical trajectories rejected due to poor testing performance compared to unrolled simulation.

Gradient Estimation and Variance Reduction in Stochastic and Deterministic Models · Cornell

Considered and rejected

Considered and rejected: Rejected direct end-to-end joint-level RL control for human-aware trajectory planning due to high dimensionality, convergence failure, and poor generalization, switching to parametric Cartesian path generation.

Effective and safe framework for human-robot interaction · IRIS - POLITO - prod

Tried and failed

second-order dynamical system imitation learning applied to robot trajectory generation. Outcome: did not generalise. Reason: unaligned initial velocities cause severe trajectory deviations without directional dissipation

Geometric Learning: Leveraging differential geometry for learning and control · EPFL

Considered and rejected

Considered and rejected: Rejected conditioning policies on goal images or continuous trajectory sketches due to over-specification, high input dimensionality, and lack of orientation context.

Scaling robot learning with heterogeneous data from the real world, simulation, and the web · UT Austin

Tried and failed

embedding neural networks directly in quadratic programs applied to real-time robot trajectory optimization. Outcome: worse than baseline. Reason: frequent optimization failures compared to using Gaussian processes inside the quadratic program

Advancing the Adaptability of Compliant Robot Controllers for Exploration, Interaction, and Manipulation · EPFL

Tried and failed

joint multi-waypoint policy optimization with RL applied to trajectory synthesis for robotic manipulation. Outcome: unstable. Reason: optimizing all trajectory waypoints simultaneously caused performance degradation over training time

Enhancing Capabilities of Assistive Robotic Arms: Learning, Control, and Object Manipulation · Virginia Tech

Tried and failed

Taylor series expansion of neural network cost applied to iterative linear quadratic regulator trajectory optimization. Outcome: worse than baseline. Reason: local quadratic approximations of deep discriminators are less effective than explicitly structured quadratic compositions

Imitation learning from observation · UT Austin

Tried and failed

Direct open-loop execution of extracted trajectory priors applied to robotic grasping and manipulation tasks. Outcome: did not generalise. Reason: Demonstration trajectories without interactive exploration cannot handle real-world physical and kinematic discrepancies.

Teaching Robots using Interactive Imitation Learning · Virginia Tech

Curse of dimensionality and computational complexity cause planning intractability

11 theses · 7 institutions

Expanding planning horizons, evaluating large scenario trees, or adding optimization parameters significantly increases computation time and exceeds real-time execution budgets. As a result, dense forward sampling, reachability updates, and dynamic programming formulations become computationally prohibitive or exceed memory and mesh bounds.

Tried and failed

joint optimization of boundary points and time allocations applied to multi-segment trajectory planning. Outcome: worse than baseline. Reason: tripling optimization dimensionality degraded convergence and solution quality for fixed iteration budgets

Multi-fidelity Optimal Trajectory Generation: Optimal Experiment Design for Robot Learning · MIT

Tried and failed

increasing parameterization dimensionality in evolutionary trajectory optimization applied to continuous path planning in vector fields. Outcome: worse than baseline. Reason: Expanded search space degraded genetic algorithm optimization efficiency compared to lower-dimensional representations.

EXPLOITING FLOWS FOR ORIENTEERING AND PLANNING PROBLEMS · Penn

Considered and rejected

Considered and rejected: Rejected dynamic programming due to the curse of dimensionality on minimum-time trajectory problems.

Artificial Drivers for Online Time-Optimal Vehicle Trajectory Planning and Control · IRIS - UNITN - prod

Considered and rejected

Considered and rejected: Rejected full dynamic programming / Q-table pre-computation for multi-task navigation due to excessive real-time human cognitive load and curse of dimensionality.

A modular attention hypothesis for modeling visuomotor behaviors · UT Austin

Tried and failed

coupled zonotope reachability model predictive control applied to real-time robot trajectory planning. Outcome: too slow. Reason: computational complexity was two orders of magnitude too high to meet real-time update frequency requirements

Safe Bipedal Locomotion and Navigation in Uncertain Environments · Georgia Tech

Tried and failed

scenario-tree min-max model predictive control applied to real-time legged robot trajectory planning. Outcome: too slow. Reason: exponential growth of uncertainty realizations across the planning horizon made real-time optimization computationally intractable

Robust Predictive Control for Legged Locomotion · Virginia Tech

Considered and rejected

Considered and rejected: Rejected sampling-based trajectory planners (MCTS/RRT) due to prohibitive computational expense when repeatedly evaluating counterfactual belief states.

Enabling Human-Multi-Robot Collaborative Visual Exploration in Underwater Environments · MIT

Considered and rejected

Considered and rejected: Rejected direct non-iterative dense forward sampling (M >= 100,000) due to prohibitive compute when the optimal trajectory distribution is initially unknown.

Feynman-Kac Numerical Techniques for Stochastic Optimal Control · Georgia Tech

Considered and rejected

Considered and rejected: Rejected distributionally robust optimization (DRO) and brute-force search for trajectory planning due to severe computational complexity and overly conservative trajectories.

Dynamic Connected Automated Vehicle Trajectory and Traffic Signal Timing Optimization · Virginia Tech

Tried and failed

gradient descent trajectory optimization applied to orbital maneuver estimation. Outcome: too slow. Reason: excessive runtime from repeated forward propagations and inability to directly provide uncertainty quantification

Applications of random finite set-based multi-target trackers in space situational awareness · UT Austin

Lost to a baseline

Normal h mesh refinement method failed to solve the low-thrust trajectory problem (Ex 6.6.3) within the 4000 maximum mesh points limit (where ph adaptive mesh refinement succeeded).

Development of Parallel Indirect Methods for Solving Constrained Optimal Control Problems · ResearchWorks

Neglecting full-body dynamics and kinematic limits produces physically infeasible trajectories

11 theses · 8 institutions

Formulating trajectory planners without accounting for joint torque bounds, actuator limits, or terrain resistance results in kinematically unachievable paths and physical foot slipping. Relying on simplified kinematic approximations or underconstrained proximal joints causes severe kinetic errors, configuration singularities, and uncontrolled joint drift.

Tried and failed

greedy viewpoint selection without kinematic constraints applied to autonomous robotic coverage path planning. Outcome: worse than baseline. Reason: ignoring reachability and velocity bounds generates dynamically infeasible trajectories, reducing actual coverage

Enhancing viewpoint planning and time-optimal trajectory generation for autonomous robot surveying · UT Austin

Tried and failed

shortest distance path planning applied to off-road vehicle trajectory optimization. Outcome: worse than baseline. Reason: minimizing geometric distance ignores terrain motion resistance, causing highest energy loss and lowest tractive efficiency

Evaluation of off-road uninhabited ground vehicle mobility using discrete element method and scalability investigation · Cranfield

Tried and failed

heuristic trajectory generation decoupling dynamics from planning applied to bipedal locomotion over large obstacles. Outcome: unstable. Reason: heuristic swing trajectories ignored full-body dynamics, causing foot slip on high steps

Real-Time Perception and Mixed-Integer Footstep Control for Underactuated Bipedal Walking on Rough Terrain · Penn

Tried and failed

trajectory optimization with non-zero initial boundary curvature applied to undulatory locomotion gait generation. Outcome: worse than baseline. Reason: enforced non-zero boundary curvature produced energetically suboptimal intermediate S-shaped body configurations

Gait optimality for undulatory locomotion with applications to C. elegans phenotyping · Imperial

Tried and failed

trajectory optimization assuming lower-tier subsystem kinematics applied to multi-joint prosthetic limb design. Outcome: did not generalise. Reason: optimizing for distal kinematics caused kinetic errors and asymmetric gait when knee joint dynamics were missing

Integrated Prosthetic Leg Design Frameworks for People with an Above-Knee Amputation · MIT

Tried and failed

centroidal dynamics with full kinematics trajectory optimization applied to humanoid dynamic jumping maneuvers. Reason: omits joint torque and electrical power limits, yielding physically infeasible trajectories

A Model-Based Planning and Control Framework for Parkour-Style Legged Locomotion · MIT

Tried and failed

trajectory optimization with underconstrained proximal joints applied to multibody dynamic motion prediction. Reason: omitting realistic dynamic torso and shoulder rotation constraints caused downstream distal trajectory prediction errors

Optimization-Based Motion Prediction for Tossing Tasks · Texas Tech

Considered and rejected

Considered and rejected: Rejected directly optimizing feature vectors followed by trajectory reconstruction in motion attribution because target feature vectors often have no physically realizable trajectory satisfying domain constraints.

Making Robot Behaviors Automatically Transparent · ResearchWorks

Tried and failed

differential flatness mapping applied to aggressive quadrotor trajectory generation. Reason: mapping singularities near 90-degree tilts from hover restrict aggressive acceleration and pitch/roll trajectories

Trajectory Planning for Flights in Multiagent and Dynamic Environments · MIT

Tried and failed

simplified empirical component models in trajectory optimization applied to aircraft trajectory optimization. Reason: Oversimplified variable dependencies failed to meet required optimization accuracy standards.

Multidisciplinary methods for evaluation and optimisation of aircraft flight performance. PhD in Aerospace · Cranfield

Tried and failed

Kinematic control without null-space optimization applied to Redundant manipulator trajectory tracking. Reason: Unmanaged redundancy caused primary joints to drift and exceed physical joint limits during continuous motion.

Robotic manipulation of hand tools: estimation, planning, control and finger gaiting · Iowa State

Trajectory optimizers become trapped in local minima or fail to converge in complex environments

8 theses · 6 institutions

Trajectory optimization and sampling methods frequently diverge or terminate prematurely when navigating tight passages and non-convex obstacle fields. Flawed initial guesses, indefinite Hessian terms, and unimodal sampling distributions leave algorithms stuck in sub-optimal local minima.

Tried and failed

semi-analytic initial guess for trajectory optimization applied to low-thrust orbit insertion trajectory design. Outcome: did not converge. Reason: insufficient control flexibility per revolution combined with high-dimensional nonlinear programming design space

A Shape Based Approach for Preliminary Design of Low Thrust Space Trajectories · Iowa State

Tried and failed

sampling-based model predictive control applied to high-dimensional trajectory planning around obstacles. Outcome: did not converge. Reason: Unimodal trajectory sampling became trapped in sub-optimal local minima between obstacles in high-dimensional spaces

Of Priors and Particles: Structured and Distributed Approaches to Robot Perception and Control · Georgia Tech

Tried and failed

dynamic movement primitives under waypoint constraints applied to robot trajectory generation with obstacle avoidance. Outcome: did not converge. Reason: hyperparameters failed to reliably find trajectories satisfying both intermediate via-point and obstacle avoidance constraints

Automatically Encoding, Modifying, and Finding Robot Skills to Repair High-Level Tasks · Cornell

Tried and failed

direct parameter sampling for trajectory optimization applied to articulated robotic manipulation planning. Outcome: did not converge. Reason: sampling planners frequently terminated prematurely returning incomplete approximate solutions instead of valid paths

Remote robotic manipulation task execution using affordance primitives · UT Austin

Tried and failed

local velocity obstacle trajectory planning applied to multi-agent obstacle avoidance. Outcome: did not converge. Reason: Local optimization gets trapped in local minima formed by static obstacles without global path lookahead

Conceptual-Level Analysis and Design of Unmanned Air Traffic Management Systems · Georgia Tech

Tried and failed

sequential path stepping gradient descent optimization applied to constrained robotic trajectory planning. Outcome: did not converge. Reason: algorithm failed in environments with tight obstacles and long paths

Remote robotic manipulation task execution using affordance primitives · UT Austin

Tried and failed

naive penalty differential dynamic programming applied to constrained non-convex trajectory optimization. Outcome: did not converge. Reason: indefinite Hessian terms in non-convex obstacle fields and tight environments cause local minima or divergence

Safety Embedded Optimal Decision Making and Control Via Barrier States · Georgia Tech

Tried and failed

semi-analytic initial guesses for high-revolution trajectory optimization applied to many-revolution low-thrust trajectory optimization. Outcome: did not converge. Reason: linearized/semi-analytic approximations accumulated too much error over many revolutions for the numerical optimizer to converge

A Shape Based Approach for Preliminary Design of Low Thrust Space Trajectories · Iowa State

Lost to a baseline

On the CSAIL dataset, DC-SAM attained higher trajectory error than graduated nonconvexity (GNC) due to entrapment in local optima.

Lifelong, learning-augmented robot navigation · Woods Hole

Considered and rejected

Considered and rejected: Rejected nonlinear trajectory optimization due to heavy initialization dependence and lack of convergence guarantees

An Optimization Approach to Certified Manipulation · MIT

Execution delays, sensor drift, and control approximations induce tracking instability and oscillations

9 theses · 5 institutions

Outdated state estimates from computational latency, unmodeled sensor drift, and low navigation gains lead to limit cycle oscillations and divergence. Dynamic weight adjustments and linearized reduced-order models struggle during aggressive velocity changes, causing severe trajectory tracking errors and whiplash effects.

Tried and failed

probabilistic movement primitives from full trajectory demonstrations applied to dynamic robot handover tasks. Outcome: unstable. Reason: generated abrupt, non-smooth motions causing visual tracking failure on wrist cameras

A Stochastic Approach to Contact-rich Manipulation · EPFL

Tried and failed

torque-constrained pseudoinverse redundancy resolution applied to redundant robotic manipulator control. Outcome: unstable. Reason: dynamic weighting modifications caused whiplash instabilities along extended trajectory tracking

Systems design and control of a freeflying space robotic manipulator system (ATLAS) for in-orbit satellite servicing operations · Cranfield

Tried and failed

model predictive control with order-reduced linearized models applied to dynamic visual trajectory tracking. Outcome: did not generalise. Reason: rapid operating-point velocity changes and cueing time delays degraded linear model validity

Model-Based Life Extending Control for Rotorcraft · Georgia Tech

Tried and failed

dead-reckoning inertial navigation applied to underwater vehicle localization. Outcome: unstable. Reason: unbounded integration error caused by sensor drift over time

Human-Interactions with Robotic Cyber-Physical Systems (CPS) for Facilitating Construction Progress Monitoring · Virginia Tech

Tried and failed

reproducing kernel hilbert space adaptive control applied to autonomous underwater vehicle trajectory tracking. Outcome: unstable. Reason: baseline non-parametric adaptive control exhibited steady-state limit cycle oscillations in pitch and depth

Data-Driven, Non-Parametric Model Reference Adaptive Control Methods for Autonomous Underwater Vehicles · Virginia Tech

Tried and failed

proportional navigation with low navigation gain applied to autonomous vehicle terminal guidance. Outcome: unstable. Reason: navigation gain below theoretical limit caused unbounded acceleration commands near target interception

Analytical and experimental study of a recovery guidance system using vortex lattice method, output regulator and proportional navigation guidance · Iowa State

Tried and failed

open-loop waypoint passing without delay compensation applied to real-time vehicle trajectory planning. Outcome: unstable. Reason: Vehicle movement during planning computation rendered initial pose outdated, causing erratic trajectory execution

Terrain aware tactical motion planning and control algorithms for off-road UGVs in GNSS denied hostile environments · Virginia Tech

Tried and failed

egocentric visual speed guidance cueing applied to manual closed-loop vehicle trajectory tracking. Outcome: unstable. Reason: velocity tracking errors integrated into unrecoverable position lag, driving pilot-induced oscillations in low visibility

Augmented reality cueing methodologies for rotorcraft shipboard landings · Georgia Tech

Lost to a baseline

Weighted pseudoinverse and GUAM baseline diverged and exceeded trajectory tracking error bounds (±2 m/s / ±2 deg/s) during transition-to-cruise switch under n1 and delta_aL failure.

Simulation and Flight-Test Evaluation of Fault-Tolerant Control Allocation Strategies for eVTOL Aircraft · Virginia Tech

Overly conservative safety margins and aggressive pruning eliminate feasible collision-free paths

7 theses · 7 institutions

Applying oversized geometric hulls, high risk thresholds, and conservative swept-volume approximations unduly shrinks the drivable space and rejects valid trajectories. Furthermore, pruning rollouts based strictly on destination proximity discards necessary temporary detours required to clear congested obstacles.

Tried and failed

obstacle avoidance without proximity scaling applied to dynamic robot trajectory planning. Outcome: worse than baseline. Reason: omitting proximity scaling caused premature, overly conservative avoidance and unnecessary trajectory detours

thesis in the field of Mechanical Engineering: Relevance for Human-Robot Collaboration: Definitions, Systems, Algorithms, and Applications · MIT

Tried and failed

Goal-proximity rollout pruning for dynamic obstacle prediction applied to multi-agent collision avoidance trajectory planning. Outcome: worse than baseline. Reason: Avoiding congestion often requires agents to temporarily move away from their destinations, which pruned valid paths.

Applying FastMDP to complex aerospace-related problems · Iowa State

Tried and failed

bidirectional sampling-based motion planning applied to manipulation trajectory planning in dense clutter. Reason: Cannot find paths when strictly collision-free trajectories do not exist, triggering destructive fallback behaviors

Robotic manipulation in clutter with object-level semantic mapping · Imperial

Lost to a baseline

Arbitrary polyhedral decomposition produced narrower corridors and slower, less rounded trajectories than point cloud ellipsoidal inflation on the Astrobee testbed.

Geometric Methods For The Planning, Control, And Estimation Of Free-Flying Autonomous Systems · Penn

Lost to a baseline

Unimodal projection baseline achieved lower Euclidean distance to the original trajectory than multimodal projection on roundabout trajectories because its oversized hull did not penalize shortcutting across non-drivable zones.

Strategic decision making for multi-agent interaction : a study in game theory and optimization · UT Austin

Tried and failed

stochastic reachability analysis with high risk threshold applied to collision avoidance trajectory planning. Reason: overly restrictive probability threshold truncated reachable sets, omitting critical obstacle occupancy states

Threat Assessment and Proactive Decision-Making for Crash Avoidance in Autonomous Vehicles · Virginia Tech

Tried and failed

swept volume over-approximation for collision avoidance applied to trajectory and manipulator structure synthesis. Reason: over-approximation was too conservative, eliminating feasible paths in constrained obstacle navigation tasks

Task-Based Design Synthesis Of Modular Manipulators · Cornell

Constant-motion assumptions fail to handle dynamic obstacles and interactive agents

7 theses · 6 institutions

Trajectory planners that assume static obstacles, zero lead-vehicle acceleration, or constant obstacle velocities trigger collisions when other agents maneuver. Failing to track moving obstacles or dynamic environmental flow fields causes excessive control lag, string instability, and severe trajectory detours.

Tried and failed

deterministic dynamic game planning without uncertainty tracking applied to multi-agent interactive robot navigation. Outcome: worse than baseline. Reason: failed to account for motion noise and heterogeneous observation uncertainty during aggressive maneuvers

Learning and control for interactions in mixed human-robot environments · MIT

Tried and failed

velocity obstacles with large planning horizons applied to robot navigation in dynamic crowds. Outcome: worse than baseline. Reason: over-constraining constant acceleration assumptions caused excessive collisions and inefficiency

Geometric and Learning Methods for Robots to Navigate in Human Crowds with Application to Smart Mobility Devices · EPFL

Lost to a baseline

Robot-only handover planner achieved a 62% success rate on randomized obstacle trajectories compared to 57% for the collaborative MPC optimizer when partners were uncooperative

The Expert is the Obstacle: Building a General Framework for Learned Robot Motion · ResearchWorks

Considered and rejected

Considered and rejected: Rejected estimating MPC lead vehicle trajectory assuming zero acceleration, because inaccurate prediction degrades follower string stability and causes lagging/overshooting responses

Longitudinal Control for Self-driving Cars with Traffic Flow Considerations: Theory, Design, and Experiments · Georgia Tech

Considered and rejected

Considered and rejected: Rejected deterministic variational least-squares trajectory recovery due to ill-posedness and high-dimensional multi-modal optimization landscapes caused by agent symmetry

MULTI-AGENT SYSTEMS: OBSERVABILITY, CLASSIFICATION AND CLUSTERING PREDICTION · JScholarship

Tried and failed

static geometric path planning on time-varying vector fields applied to energy-optimal vehicle navigation in dynamic currents. Outcome: did not generalise. Reason: Paths optimized for a single time snapshot waste energy fighting opposing flows when currents shift over time

Energy efficient path planning and model checking for long endurance unmanned surface vehicles. · Cranfield

Tried and failed

joint optimization of control barrier functions and policy applied to safe robot navigation with dynamic obstacles. Outcome: worse than baseline. Reason: inability to handle moving targets leading to high collision rates

Personalized, Safe, and Interactive Robot Programming via Human Demonstrations · Georgia Tech

Left open by the authors

Problems the authors named and did not get to.

Left open

Benchmark DDPG, TD3, PPO, DDP, collocation, and MPPI against SAC for learning maneuver automaton motion primitives in navigation simulations. Blocker: None

Autonomous Methods for Learning and Pruning Motion Primitives for Navigation and Adversarial Tasks · Georgia Tech

Left open

Train machine learning models (Random Forest, SVM, Deep Learning) on LiDAR trajectory data to classify driver behavior instead of rule-based kinematic thresholds. Blocker: Access to the thesis author's roadside LiDAR trajectory dataset and proprietary VISSIM simulation environment.

Calibration of Microscopic Traffic Simulation Models for Proactive Safety Performance based on LiDAR Trajectory Data · Texas Tech

Left open

Extend adversarial trajectory attack methods to incorporate collision velocity and group behaviors for autonomous driving motion forecasting models. Blocker: None

Deep Generative Models for Autonomous Driving: from Motion Forecasting to Realistic Image Synthesis · EPFL

Left open

Extend perception-aware view optimization and trajectory planning to mobile vehicles equipped with robotic manipulators in simulation. Blocker: None

Perception-aware planning for differentially flat robots · MIT

Left open

Integrate symbolic task-level planning with low-level continuous motor trajectory optimization and reactive re-planning. Blocker: The goal is a high-level research direction without concrete benchmarks, architectures, or targets specified.

Learning to act with objects, relations and physics · MIT

Left open

Integrate dynamic stability and 3D force/torque constraints into adaptive-dimensionality planning for humanoid navigation on non-horizontal terrain. Blocker: None

Planning With Adaptive Dimensionality · Penn

Left open

Design non-linear interpolation paths for flow matching to incorporate non-Euclidean geometries or trajectory constraints. Blocker: High ambiguity regarding specific manifolds, constraint types, and concrete mathematical formulations to test

Generative modeling by interpolating between distributions · UT Austin

Left open

Extend the decoupled RRT and direct collocation trajectory optimization formulation from a 2D model to a full 3D quadruped system. Blocker: None

Decoupled Kinodynamic Planning for a Quadruped Robot over Complex Terrain · MIT

Left open

Extend cross-learning to navigation problems featuring complex dynamics from human-machine interaction. Blocker: Vague task direction lacking specific formulation, targets, or interactive dynamics models

Graph Machine Learning Under Requirements · Penn

Left open

Apply the DMD/EDMD Koopman framework to motion capture trajectories of bipedal animals such as kangaroos and ostriches. Blocker: Requires specialized motion capture data or live animal laboratory facilities for kangaroos and ostriches.

Animal Motion Analysis and Approximation for Robotics · Virginia Tech

Checking a claim in this area?

We can run the same search on any method or claim. If nothing turns up, we will say so, and that proves nothing on its own.